m
Game Developers
Science, Math, a nd Code for Realistic Effects
O’REILLY
www.it-ebooks.info
David M. Bourg
& Bryan By wale c
Game Programming
Physics for Game Developers
If you want to enrich your game’s experience with physics-based
realism, the expanded edition of this classic book details physics
principles applicable to game development. You’ll learn about
collisions, explosions, sound, projectiles, and other effects used in
games on Wii, PlayStation, Xbox, smartphones, and tablets. You’ll
also get a handle on how to take advantage of various sensors such
as accelerometers and optical tracking devices.
Authors David Bourg and Bryan Bywalec show you how to develop
your own solutions to a variety of problems by providing technical
background, formulas, and a few code examples. This updated
book is indispensable whether you work alone or as part of a team.
■ Refresh your knowledge of classical mechanics, including
kinematics, force, kinetics, and collision response
■ Explore rigid body dynamics, using real-time 2D and 3D
simulations to handle rotation and inertia
■ Apply concepts to real-world problems: model the behavior of
boats, airplanes, cars, and sports balls
■ Enhance your games with digital physics, using
accelerometers, touch screens, GPS, optical tracking devices,
and 3D displays
■ Capture 3D sound effects with the OpenAL audio API
“Physics for Game
Developers has a wealth
of information based
on real world physics
problems that is
immediately usable
in game code.”
—Paul Zirkle
Lead Games Engineer
at Disney Interactive
David Bourg, owner of MiNO Marine—a Naval architecture and
marine services firm—also formed a company in the 1990s that
developed children’s games, casino games, and various PC to Mac
ports. He’s the co-author of AIfor Game Programmers (O'Reilly).
Bryan Bywalec is an architect at MiNO Marine, where accurate
simulation of the physical world is necessary on a daily basis. In
his passion for physics, he enjoys modding games (like Kerble
Space Program) that places physics on center stage.
US$44.99 CAN $47.99
ISBN: 978-1-449-39251-2
Twitter: @oreillymedia
facebook.com/oreilly
O’REILLY®
oreilly.com
www.it-ebooks.info
SECOND EDITION
Physics for Game Developers
David M. Bourg and Bryan Bywalec
O’REILLY®
Beijing • Cambridge • Farnham • Koln • Sebastopol • Tokyo
www.it-ebooks.info
Physics for Game Developers, Second Edition
by David M. Bourg and Bryan Bywalec
Copyright © 2013 David M. Bourg and Bryan Bywalec. All rights reserved.
Printed in the United States of America.
Published by O’Reilly Media, Inc., 1005 Gravenstein Highway North, Sebastopol, CA 95472.
O’Reilly books may be purchased for educational, business, or sales promotional use. Online editions are
also available for most titles (http://my.safaribooksonline. com). For more information, contact our corporate/
institutional sales department: 800-998-9938 or corporate@oreilly.com.
Editors: Andy Oram and Rachel Roumeliotis Indexer: Lucie Haskins
Production Editor: Christopher Hearse Cover Designer: Randy Comer
Copyeditor: Rachel Monaghan Interior Designer: David Futato
Proofreader: Amanda Kersey Illustrator: Rebecca Demarest
April 2013: Second Edition
Revision History for the Second Edition:
2013-04-09: First release
See http://oreilly.com/catalog/errata.csp?isbn=9781449392512 for release details.
Nutshell Handbook, the Nutshell Handbook logo, and the O’Reilly logo are registered trademarks of O’Reilly
Media, Inc. Physics for Game Developers, 2nd Edition, the image of a cat and mouse, and related trade dress
are trademarks of O’Reilly Media, Inc.
Many of the designations used by manufacturers and sellers to distinguish their products are claimed as
trademarks. Where those designations appear in this book, and O’Reilly Media, Inc., was aware of a trade¬
mark claim, the designations have been printed in caps or initial caps.
While every precaution has been taken in the preparation of this book, the publisher and authors assume
no responsibility for errors or omissions, or for damages resulting from the use of the information contained
herein.
ISBN: 978-1-449-39251-2
[LSI]
www.it-ebooks.info
Preface.
Table of Contents
. xi
Part 1. Fundamentals
1. Basic Concepts.
. 3
Newton’s Laws of Motion
3
Units and Measures
4
Coordinate System
6
Vectors
7
Derivatives and Integrals
8
Mass, Center of Mass, and Moment of Inertia
9
Newton’s Second Law of Motion
20
Inertia Tensor
24
Relativistic Time
29
2. Kinematics.
. 35
Velocity and Acceleration
36
Constant Acceleration
39
Nonconstant Acceleration
41
2D Particle Kinematics
42
3D Particle Kinematics
45
X Components
46
Y Components
47
Z Components
48
The Vectors
48
Hitting the Target
49
Kinematic Particle Explosion
54
Rigid-Body Kinematics
61
Local Coordinate Axes
62
iii
www.it-ebooks.info
Angular Velocity and Acceleration
62
Force.
. 71
Forces
71
Force Fields
72
Friction
73
Fluid Dynamic Drag
75
Pressure
76
Buoyancy
77
Springs and Dampers
79
Force and Torque
80
Summary
83
Kinetics.
. 85
Particle Kinetics in 2D
87
Particle Kinetics in 3D
91
X Components
94
Y Components
95
Z Components
95
Cannon Revised
95
Rigid-Body Kinetics
99
Collisions.
.103
Impulse-Momentum Principle
104
Impact
105
Linear and Angular Impulse
112
Friction
115
Projectiles.
.119
Simple Trajectories
120
Drag
124
Magnus Effect
132
Variable Mass
138
Part II. Rigid-Body Dynamics
7. Real-Time Simulations. 143
Integrating the Equations of Motion 144
Euler’s Method 146
Better Methods 153
iv | Table of Contents
www.it-ebooks.info
Summary
159
8. Particles. 161
Simple Particle Model 166
Integrator 169
Rendering 170
The Basic Simulator 170
Implementing External Forces 172
Implementing Collisions 175
Particle-to-Ground Collisions 175
Particle-to-Obstacle Collisions 181
Tuning 186
9. 2D Rigid-Body Simulator. 189
Model 190
Transforming Coordinates 197
Integrator 198
Rendering 200
The Basic Simulator 201
Tuning 204
10. Implementing Collision Response.205
Linear Collision Response 206
Angular Effects 213
11. Rotation in 3D Rigid-Body Simulators. 227
Rotation Matrices 228
Quaternions 232
Quaternion Operations 234
Quaternions in 3D Simulators 239
12. 3D Rigid-Body Simulator. 243
Model 243
Integration 247
Flight Controls 250
13. Connecting Objects. 255
Springs and Dampers 257
Connecting Particles 258
Rope 258
Connecting Rigid Bodies 265
Links 265
Table of Contents | v
www.it-ebooks.info
Rotational Restraint
275
Physics Engines.
. 281
Building Your Own Physics Engine
281
Physics Models
283
Simulated Objects Manager
284
Collision Detection
285
Collision Response
286
Force Effectors
287
Numerical Integrator
288
Part III. Physical Modeling
Aircraft.
. 293
Geometry
294
Lift and Drag
297
Other Forces
302
Control
303
Modeling
305
Ships and Boats.
. 321
Stability and Sinking
323
Stability
323
Sinking
325
Ship Motions
326
Heave
327
Roll
327
Pitch
328
Coupled Motions
328
Resistance and Propulsion
328
General Resistance
328
Propulsion
334
Maneuverability
335
Rudders and Thrust Vectoring
336
Cars and Hovercraft.
.339
Cars
339
Resistance
339
Power
340
Stopping Distance
341
Steering
342
vi | Table of Contents
www.it-ebooks.info
Hovercraft
345
How Hovercraft Work
345
Resistance
347
Steering
350
Guns and Explosions.
. 353
Projectile Motion
353
Taking Aim
355
Zeroing the Sights
357
Breathing and Body Position
360
Recoil and Impact
361
Explosions
362
Particle Explosions
363
Polygon Explosions
366
Sports.
. 369
Modeling a Golf Swing
370
Solving the Golf Swing Equations
373
Billiards
378
Implementation
380
Initialization
383
Stepping the Simulation
386
Calculating Forces
388
Handling Collisions
393
Part IV. Digital Physics
20. Touch Screens.403
Types of Touch Screens 403
Resistive 403
Capacitive 404
Infrared and Optical Imaging 404
Exotic: Dispersive Signal and Surface Acoustic Wave 404
Step-by-Step Physics 404
Resistive Touch Screens 404
Capacitive Touch Screens 408
Example Program 410
Multitouch 410
Other Considerations 411
Haptic Feedback 411
Modeling Touch Screens in Games 411
Table of Contents | vii
www.it-ebooks.info
Difference from Mouse-Based Input 412
Custom Gestures 412
21. Accelerometers. 413
Accelerometer Theory 414
MEMS Accelerometers 416
Common Accelerometer Specifications 417
Data Clipping 417
Sensing Orientation 418
Sensing Tilt 420
Using Tilt to Control a Sprite 420
Two Degrees of Freedom 421
22. Gaming from One Place to Another. 427
Location-Based Gaming 427
Geocaching and Reverse Geocaching 428
Mixed Reality 428
Street Games 428
What Time Is It? 429
Two-Dimensional Mathematical Treatment 429
Location, Location, Location 433
Distance 433
Great-Circle Heading 435
Rhumb Line 436
23. Pressure Sensors and Load Cells. 439
Under Pressure 440
Example Effects of High Pressure 440
Button Mashing 442
Load Cells 444
Barometers 448
24. 3D Display.451
Binocular Vision 451
Stereoscopic Basics 454
The Left and Right Frustums 454
Types of Display 458
Complementary-Color Anaglyphs 458
Linear and Circular Polarization 459
Liquid-Crystal Plasma 462
Autostereoscopy 463
Advanced Technologies 465
viii | Table of Contents
www.it-ebooks.info
Programming Considerations 467
Active Stereoization 467
Passive Stereoization 469
25. Optical Tracking. 471
Sensors and SDKs 472
Kinect 472
OpenCV 473
Numerical Differentiation 474
26. Sound. 477
What Is Sound? 477
Characteristics of and Behavior of Sound Waves 481
Harmonic Wave 481
Superposition 483
Speed of Sound 484
Attenuation 485
Reflection 486
Doppler Effect 488
3D Sound 489
How We Hear in 3D 489
A Simple Example 491
A. Vector Operations. 495
B. Matrix Operations. 507
C. Quaternion Operations. 517
Bibliography. 529
Index. 535
Table of Contents | ix
www.it-ebooks.info
www.it-ebooks.info
Preface
Who Is This Book For?
Simply put, this book is targeted at computer game developers who do not have a strong
mechanics or physics background, charged with the task of incorporating real physics
in their games.
As a game developer, and very likely as a gamer yourself, you’ve seen products being
advertised as “ultra-realistic,” or as using “real-world physics.” At the same time you, or
perhaps your company’s marketing department, are wondering how you can spice up
your own games with such realism. Or perhaps you want to try something completely
new that requires you to explore real physics. The only problem is that you threw your
college physics text in the lake after final exams and haven’t touched the subject since.
Maybe you licensed a really cool physics engine, but you have no idea how the underlying
principles work and how they will affect what you’re trying to model. Or, perhaps you
are charged with the task of tuning someone else’s physics code but you really don’t
understand how it works. Well then, this book is for you.
Sure you could scour the Internet, trade journals, and magazines for information and
how-to’s on adding physics-based realism to your games. You could even fish out that
old physics text and start from scratch. However, you’re likely to find that either the
material is too general to be applied directly, or too advanced requiring you to search
for other sources to get up to speed on the basics. This book will pull together the
information you need and will serve as the starting point for you, the game developer,
in your effort to enrich your game’s content with physics-based realism.
This book is not a recipe book that simply gives sample code for a miscellaneous set of
problems. The Internet is full of such example programs (some very good ones we might
add). Rather than give you a collection of specific solutions to specific problems, our
aim is to arm you with a thorough and fundamental understanding of the relevant topics
such that you can formulate your own solutions to a variety of problems. We’ll do this
by explaining, in detail, the principles of physics applicable to game development, and
xi
www.it-ebooks.info
by providing complimentary hand calculation examples in addition to sample pro¬
grams.
What We Assume You Know
Although we don’t assume that you are a physics expert, we do assume that you have at
least a basic college level understanding of classical physics typical of non-physics and
non-engineering majors. It is not essential that your physics background is fresh in your
mind as the first several chapters of this book review the subjects relevant to game
physics.
We also assume that you are proficient in trigonometry, vector, and matrix math, al¬
though we do include reference material in the appendices. Further, we assume that you
have at least a basic college level understanding of calculus, including integration and
differentiation of explicit functions. Numerical integration and differentiation is a dif¬
ferent story, and we cover these techniques in detail in the later chapters of this book.
Mechanics
Most people that we’ve talked to when we was developing the concept for this book
immediately thought of flight simulators when the phrases “real physics” and “real-time
simulation” came up. Certainly cutting edge flight simulations are relevant in this con¬
text; however, many different types of games, and specific game elements, stand to
benefit from physics-based realism.
Consider this example: You’re working on the next blockbuster hunting game complete
with first-person 3D, beautiful textures, and an awesome sound track to set the mood,
but something is missing. That something is realism. Specifically, you want the game to
“feel” more real by challenging the gamer’s marksmanship, and you want to do this by
adding considerations such as distance to target, wind speed and direction, and muzzle
velocity, among others. Moreover, you don’t want to fake these elements, but rather,
you’d like to realistically model them based on the principles of physics. Gary Powell,
with MathEngine Pic, put it like this “The illusion and immersive experience of the
virtual world, so carefully built up with high polygon models, detailed textures and
advanced lighting, is so often shattered as soon as objects start to move and interact.” 1
“It’s all about interactivity and immersiveness,” says Dr. Steven Collins, CEO of Hav-
ok.com. 2 We think both these guys or right on target. Why invest so much time and
1. At the time of this book’s first edition, Gary Powell worked for MathEngine Pic. Their products included
Dynamics Toolkit 2 and Collision Toolkit 1, which handled single and multiple body dynamics. Currently
the company operates under the name CM Labs.
2. At the time of this book’s first edition, Dr. Collins was the CEO of Havok.com. Their technology handled
rigid body, soft body, cloth, and fluid and particle dynamics. Intel purchased Havok in 2005.
xii | Preface
www.it-ebooks.info
effort making your game world look as realistic as possible, but not take the extra step
to make it behave just as realistically?
Here are a few examples of specific game elements that stand to benefit, in terms of
realism, from the use of real physics:
• The trajectory of rockets and missiles including the effects of fuel burn off
• The collision of objects such as billiard balls
• The effects of gravitation between large objects such as planets and battle stations
• The stability of cars racing around tight curves
• The dynamics of boats and other waterborne vehicles
• The flight path of a baseball after being struck by a bat
• The flight of a playing card being tossed into a hat
This is by no means an exhaustive list, but just a few examples to get you in the right
frame of mind, so to speak. Pretty much anything in your games that bounces around,
flies, rolls, slides, or isn’t sitting dead still can be realistically modeled to create com¬
pelling, believable content for your games.
So how can this realism be achieved? By using physics, of course, which brings us back
to the title of this section, the subject of mechanics. Physics is a vast field of science that
covers many different, but related subjects. The subject most applicable to realistic game
content is the subject of mechanics, which is really what’s meant by “real physics.”
By definition, mechanics is the study of bodies at rest and in motion, and of the effect
of forces on them. The subject of mechanics is subdivided into statics, which specifically
focuses on bodies at rest, and dynamics, which focuses on bodies in motion. One of the
oldest and most studied subjects of physics, the formal origins of mechanics can be
traced back more than 2000 years to Aristotle. An even earlier treatment of the subject
was formalized in Problems of Mechanics, but the origins of this work are unknown.
Although some of these early works attributed some physical phenomena to magical
elements, the contributions of such great minds as Galileo, Kepler, Euler, Lagrange,
d’Alembert, Newton, and Einstein, to name a few, have helped develop our understand¬
ing of this subject to such a degree that we have been able to achieve the remarkable
state of technological advancement that we see today.
Because you want your game content to be alive and active, we’ll primarily look at bodies
in motion and will thus delve into the details of the subject of dynamics. Within the
subject of dynamics there are even more specific subjects to investigate, namely, kine¬
matics, which focuses on the motion of bodies without regard to the forces that act on
the body, and kinetics, which considers both the motion of bodies and the forces that
act on or otherwise affect bodies in motion. We’ll take a very close look at these two
subjects throughout this book.
Preface | xiii
www.it-ebooks.info
Digital Physics
This book’s first edition focused exclusively on mechanics. More than a decade after its
release we’ve broadened our definition of game physics to include digital physics not in
the cosmological sense but in the context of the physics associated with such devices as
smart phones and their unique user interaction experience. As more platforms such as
the Wii, PlayStation, X Box and smart phones come out and are expanded developers
will have to keep up with and understand the new input and sensors technologies that
accompany these platforms in order to keep producing fresh gaming experiences. But
you shouldn’t look at this as a burden, and instead look at it as an opportunity to enhance
the user’s interactive experience with your games.
Arrangement of This Book
Physics-based realism is not new to gaming, and in fact many games on the shelves these
days advertise their physics engines. Also, many 3D modeling and animation tools have
physics engines built in to help realistically animate specific types of motion. Naturally,
there are magazine articles that appear every now and then that discuss various aspects
of physics-based game content. In parallel, but at a different level, research in the area
of real-time rigid body 3 simulation has been active for many years, and the technical
j ournals are full of papers that deal with various aspects of this subject. You’ll find papers
on subjects ranging from the simulation of multiple, connected rigid bodies to the sim¬
ulation of cloth. However, while these are fascinating subjects and valuable resources,
as we hinted earlier, many of them are of limited immediate use to the game developer
as they first require a solid understanding of the subject of mechanics requiring you to
learn the basics from other sources. Further, many of them focus primarily on the
mathematics involved in solving the equations of motion and don’t address the practical
treatment of the forces acting on the body or system being simulated.
We asked John Nagle, with Animats, what is, in his opinion, the most difficult part of
developing a physics-based simulation for games and his response was developing nu¬
merically stable, robust code. 4 Gary Powell echoed this when he told me that minimizing
the amount of parameter tuning to produce stable, realistic behavior was one of the
most difficult challenges. We agree; speed and robustness in dealing with the mathe¬
matics of bodies in motion are crucial elements of a simulator. And on top of that, so
are completeness and accuracy in representing the interacting forces that initiate and
3. A rigid body is formally defined as a body, composed of a system of particles, whose particles remain at fixed
distances from each other with no relative translation or rotation among particles. Although the subject of
mechanics deals with flexible bodies and even fluids such as water, we’ll focus our attention on bodies that
are rigid.
4. At the time of this book’s first edition, John Nagle was the developer of Falling Bodies, a dynamics plug-in
for Softimage13D.
xiv | Preface
www.it-ebooks.info
perpetuate the simulation in the first place. As you’ll see later in this book, forces govern
the behavior of objects in your simulation and you need to model them accurately if
your objects are to behave realistically.
This prerequisite understanding of mechanics and the real world nature of forces that
may act on a particular body or system have governed the organization of this book.
Generally, this book is organized in four parts with each building on the material covered
in previous parts:
Part I, Fundamentals
A mechanics refresher, comprising Chapters 1 through 6.
Chapter 1, Basic Concepts
This warm up chapter covers the most basic of principles that are used and referred
to throughout this book. The specific topics addressed include mass and center of
mass, Newtons Laws, inertia, units and measures, and vectors.
Chapter 2, Kinematics
This chapter covers such topics as linear and angular velocity, acceleration, mo¬
mentum, and the general motion of particles and rigid bodies in two and three
dimensions.
Chapter 3, Force
The principles of force and torque are covered in this chapter, which serves as a
bridge from the subject of kinematics to that of kinetics. General categories of forces
are discussed including drag forces, force fields, and pressure.
Chapter 4, Kinetics
This chapter combines elements of Chapters 2 and 3 to address the subject of ki¬
netics and explains the difference between kinematics and kinetics. Further dis¬
cussion treats the kinetics of particles and rigid bodies in two and three dimensions.
Chapter 5, Collisions
In this chapter we’ll cover particle and rigid body collision response, that is, what
happens after two objects run in to each other.
Chapter 6, Projectiles
This chapter will focus on the physics of simple projectiles laying the ground work
for further specific modeling treatment in later chapters.
Part II, Rigid-Body Dynamics
An introduction to real time simulations, comprising Chapters 7 through 14.
Chapter 7, Real-Time Simulations
This chapter will introduce real-time simulations and detail the core of such sim¬
ulations—the numerical integrator. Various methods will be presented and cover¬
age will include stability and tuning.
Preface | xv
www.it-ebooks.info
Chapter 8, Particles
Before diving into rigid body simulations, this chapter will show how to implement
a particle simulation, which will be extended in the next chapter to include rigid
bodies.
Chapter 9, 2D Rigid-Body Simulator
This chapter will extend the particle simulator from the previous chapter showing
how to implement rigid bodies, which primarily consists of adding rotation and
dealing with the inertia tensor.
Chapter 10, Implementing Collision Response
Collision detection and response will be combined to implement real-time collision
capabilities in the 2D simulator.
Chapter 11, Rotation in 3D Rigid-Body Simulators
This chapter will address how to handle rigid body rotation in 3D including how
to deal with the inertia tensor. Then we’ll show the reader how to extend the 2D
simulator to 3D.
Chapter 12, 3D Rigid-Body Simulator
Multiple unconnected bodies will be incorporated in the simulator in this chapter.
Introduction of multiple bodies requires resolution of multiple rigid body colli¬
sions, which can be very tricky. Issues of stability and realism will be covered.
Chapter 13, Connecting Objects
Taking things a step further, this chapter will show how to join rigid bodies forming
connected bodies, which may be used to simulate human bodies, complex vehicles
that may blow apart, among many other game objects. Various connector types will
be considered.
Chapter 14, Physics Engines
In this chapter, specific aspects of automobile performance are addressed, including
aerodynamic drag, rolling resistance, skidding distance, and roadway banking.
Part III, Physical Modeling
A look at some real world problems, comprising Chapters 15 through 19.
Chapter 15, Aircraft
This chapter focuses on the elements of flight including propulsor forces, drag,
geometry, mass, and most importantly lift.
Chapter 16, Ships and Boats
The fundamental elements of floating vehicles are discussed in this chapter, in¬
cluding floatation, stability, volume, drag, and speed.
Chapter 17, Cars and Hovercraft
In this chapter, specific aspects of automobile performance are addressed, including
aerodynamic drag, rolling resistance, skidding distance, and roadway banking. Ad¬
ditionally hovercraft shares some of the same characteristics of both cars and boats.
xvi | Preface
www.it-ebooks.info
This chapter will consider those characteristics that distinguish the hovercraft as a
unique vehicle. Topics covered include hovering flight, aerostatic lift, and direc¬
tional control
Chapter 18, Guns and Explosions
This chapter will focus on the physics of guns including power, recoil, and proj ectile
flight. Since we generally want things to explode when hit with a large projectile,
this chapter will also address the physics of and modeling explosions.
Chapter 19, Sports
This chapter will focus on the physics of ball sports such as baseball, golf, and tennis.
Coverage will go beyond projectile physics and include such topics as including
pitching, bat swing, bat-ball impact, golf club swing and club ball impact, plus tennis
racket swinging and racket/ball impacts.
Part IV, Digital Physics
Chapters in this part of the book will explain the physics behind accelerometers,
touch screens, GPS and other gizmos showing the reader how to leverage these
elements in their games, comprising Chapters 20 through 26.
Chapter 20, Touch Screens
Touch screens facilitate virtual tactile interfaces with mobile device games, such as
those made for the iPhone. This chapter will explain the physics of touch screen
and how the reader can leverage this interface in their games particularly with
respect to virtual physical interaction with game elements through gesturing.
Chapter 21, Accelerometers
Accelerometers are now widely used in mobile devices and game controllers al¬
lowing virtual physical interaction between players and game objects. This chapter
will explain how accelerometers work, what data they provide and how that data
can be manipulated with respect to virtual physical interaction with game elements.
Topics covered will include, but not be limited to integration of acceleration data
to derive velocities and displacements and rotations.
Chapter 22, Gaming from One Place to Another
Mobile devices commonly have GPS capabilities and this chapter will explain the
physics of the GPS system including relativistic effects. Further, GPS data will be
explained and this chapter showing the reader how to manipulate that data for
virtual interaction with game elements. For example, we’ll show the reader how to
differentiate GPS data to derive speed and acceleration among other manipulations.
Chapter 23, Pressure Sensors and Load Cells
Pressure sensing devices are used in games as a means of allowing players to interact
with game elements, for example, the Wii balance board uses pressure sensors al¬
lowing players to interact with the Wii Fit game. This chapter will explain the physics
behind such pressure sensors, what data they generate, and how to manipulate that
data for game interaction.
Preface | xvii
www.it-ebooks.info
Chapter 24, 3D Display
The new PlayStation Move and Microsoft’s Kinect use optical tracking systems to
detect the position of players’ game controllers or gestures. This chapter will explain
the physics behind optical tracking and how to leverage this technology in games.
Chapter 25, Optical Tracking
As televisions and handheld game consoles race to implement 3D displays, several
different technologies are being developed. By understanding the physics of the
glasses dependent stereoscopic displays, the new “glasses free” autostereoscopic
displays, and looking forward to holography and volumetric displays, developers
will be better positioned to leverage these effects in their games.
Chapter 26, Sound
Sound is a particularly important part of a game’s immersive experience; however,
to date no book on game physics addresses the physics of sound. This chapter will
focus on sound physics including such topics of sound speed and the Doppler Effect.
Discussions will also include why sound physics is often ignored in games, for
example, when simulating explosions in outer space.
Appendix A, Vector Operations
This appendix shows you how to implement a C++ class that captures all of the
vector operations that you’ll need to when writing 2D or 3D simulations.
Appendix B, Matrix Operations
This appendix implements a class that captures all of the operations you need to
handle 3x3 matrices.
Appendix C, Quaternion Operations
This appendix implements a class that captures all of the operations you need to
handle quaternions when writing 3D rigid body simulations.
Part I, Fundamentals focuses on fundamental topics in Newtonian mechanics such as
kinematics and kinetics. Kinematics deals with the motion of objects. We’ll cover both
linear and angular velocity and acceleration. Kinetics deals with forces and resulting
motion. Part I serves as a primer for Part II, Rigid-Body Dynamics that covers rigid
body dynamics. Readers already versed in classical mechanics can skip Part I, Funda¬
mentals without loss of continuity.
Part II, Rigid-Body Dynamics focuses on rigid body dynamics and development of both
single and multi-body simulations. This part covers numerical integration, real-time
simulation of particles and rigid bodies, and connected rigid bodies. Generally, this part
covers what most game programmers consider elements of a physics engine.
Part III, Physical Modeling focuses on physical modeling. The aim of this part is to
provide valuable physical insight for the reader so they can make better judgments on
what to include in their models and what they can safely leave out without sacrificing
physical realism. We cannot and do not attempt to cover all the possible things you
xviii | Preface
www.it-ebooks.info
might want to simulate. Instead we cover several typical things you may try to simulate
in a game such as aircraft, boats, sports balls, among others with the purpose of giving
you some insight into the physical nature of those things and some of the choices you
must make when developing suitable models.
Part IV, Digital Physics covers digital physics in a broad sense. This is an exciting topic
as it relates to the technologies associated with mobile platforms, such as smart phones
like the iPhone, and ground breaking game systems such as the Nentendo Wii. Chapters
in this part of the book will explain the physics behind accelerometers, touch screens,
GPS and other gizmos showing the reader how to leverage these elements in their games.
We recognize that these topics are not what most game programmers typically think
about when they think of game physics; however, the technologies covered play an
increasingly important role in modern mobile games and we feel it important to explain
the underlying physics behind them with the hope that you’ll be better able to leverage
these technologies in your games.
In addition to resources pertaining to real-time simulations, the Bibliography at the end
of this book will provide sources of information on mechanics, mathematics, and other
specific technical subjects, such as books on aerodynamics.
Conventions Used in This Book
The following typographical conventions are used in this book:
Constant width
Used to indicate command-line computer output, code examples, Registry keys,
and keyboard accelerators (see “Keyboard Accelerators” later in this book).
Constant width italic
Used to indicate variables in code examples.
Italic
Introduces new terms and to indicate URLs, variables, filenames and directories,
commands, and file extensions.
Bold
Indicates vector variables.
« 9 *.
This icon signifies a tip, suggestion, or general note.
This icon indicates a warning or caution.
Preface | xix
www.it-ebooks.info
We use boldface type to indicate a vector quantity, such as force, F. When referring to
the magnitude only of a vector quantity, we use standard type. For example, the mag¬
nitude of the vector force, F, is F with components along the coordinate axes, Fx, Fy,
and Fz. In the code samples throughout the book, we use the * (asterisk) to indicate
vector dot product, or scalar product, operations depending on the context, and we use
the A (caret) to indicate vector cross product.
Using Code Examples
This book is here to help you get your job done. In general, if this book includes code
examples, you may use the code in your programs and documentation. You do not need
to contact us for permission unless you’re reproducing a significant portion of the code.
For example, writing a program that uses several chunks of code from this book does
not require permission. Selling or distributing a CD-ROM of examples from O’Reilly
books does require permission. Answering a question by citing this book and quoting
example code does not require permission. Incorporating a significant amount of ex¬
ample code from this book into your product’s documentation does require permission.
We appreciate, but do not require, attribution. An attribution usually includes the title,
author, publisher, and ISBN. For example: “Physics for Game Developers, 2nd Edition
by David M. Bourg and Bryan Bywalec (O’Reilly). Copyright 2013 David M. Bourgand
Bryan Bywalec, 978-1-449-39251-2.”
If you feel your use of code examples falls outside fair use or the permission given above,
feel free to contact us at permissions@oreilly.com.
Safari® Books Online
*+ r Safari Books Online ( www.safaribooksonline.com ) is an on-demand
Ddldll digital library that delivers expert content in both book and video
form from the world’s leading authors in technology and business.
Books Online
Technology professionals, software developers, web designers, and business and crea¬
tive professionals use Safari Books Online as their primary resource for research, prob¬
lem solving, learning, and certification training.
Safari Books Online offers a range of product mixes and pricing programs for organi¬
zations, government agencies, and individuals. Subscribers have access to thousands of
books, training videos, and prepublication manuscripts in one fully searchable database
from publishers like O’Reilly Media, Prentice Hall Professional, Addison-Wesley Pro¬
fessional, Microsoft Press, Sams, Que, Peachpit Press, Focal Press, Cisco Press, John
Wiley & Sons, Syngress, Morgan Kaufmann, IBM Redbooks, Packt, Adobe Press, FT
Press, Apress, Manning, New Riders, McGraw-Hill, Jones & Bartlett, Course Technol-
xx | Preface
www.it-ebooks.info
ogy, and dozens more. For more information about Safari Books Online, please visit us
online.
How to Contact Us
Please address comments and questions concerning this book to the publisher:
O’Reilly Media, Inc.
1005 Gravenstein Highway North
Sebastopol, CA 95472
800-998-9938 (in the United States or Canada)
707-829-0515 (international or local)
707-829-0104 (fax)
We have a web page for this book, where we list errata, examples, and any additional
information. You can access this page at http://oreil.ly/Physics-GameDev2.
To comment or ask technical questions about this book, send email to bookques
tions@oreilly.com.
For more information about our books, courses, conferences, and news, see our website
at http://www.oreilly.com.
Find us on Facebook: http://facebook.com/oreilly
Follow us on Twitter: http://twitter.com/oreillymedia
Watch us on YouTube: http://www.youtube.com/oreillymedia
Preface | xxi
www.it-ebooks.info
Acknowledgments
We want to thank Andy Oram, the editor of this edition of the book, for his skillful
review of our writing and his insightful comments and suggestions, not to mention his
patience. We also want to express my appreciation to O’Reilly for agreeing to take on
this project giving us the opportunity to expand on the original edition. Furthermore,
special thanks go to all of the production and technical staff at O’Reilly.
We’d also like to thank the technical reviewers, Christian Stober and Paul Zirkle, whose
valuable insight added much to this edition.
Individually, David would like to thank his loving wife and best friend, Helena, for her
endless support and encouragement, and his wonderful daughter, Natalia, for making
every day special.
Bryan would like to thank his co-author David for the opportunity to help with the
second edition and would also like to thank his parents, Barry and Sharon, for raising
him to be curious about the world. Lastly, he would like to thank his fiancee, Anne
Hasuly, for her support without which many chapters would still be half-finished.
xxii | Preface
www.it-ebooks.info
PARTI
Fundamentals
Part I focuses on fundamental topics in Newtonian mechanics such as kinematics and
kinetics. Kinematics deals with the motion of objects; we’ll cover both linear and angular
velocity and acceleration. Kinetics deals with forces and resulting motion. Part I serves
as a primer for Part II, which covers rigid-body dynamics. Readers already versed in
classical mechanics can skip Part I without loss of continuity.
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 1
Basic Concepts
As a warm-up, this chapter will cover the most basic of the principles that will be used
and referenced throughout the remainder of this book. First, we’ll introduce Newtons
laws of motion, which are very important in the study of mechanics. Then we’ll discuss
units and measures, where we’ll explain the importance of keeping track of units in your
calculations. You’ll also have a look at the units associated with various physical quan¬
tities that you’ll be studying. After discussing units, we’ll define our general coordinate
system, which will serve as our standard frame of reference. Then we’ll explain the
concepts of mass, center of mass, and moment of inertia, and show you how to calculate
these quantities for a collection, or combination, of masses. Finally, we’ll discuss New¬
ton’s second law of motion in greater detail, take a quick look at vectors, and briefly
discuss relativistic time.
Newton's Laws of Motion
In the late 1600s (around 1687), Sir Isaac Newton put forth his philosophies on me¬
chanics in his Philosophiae Naturalis Principia Mathematica. In this work Newton sta¬
ted the now-famous laws of motion, which are summarized here:
Law I
A body tends to remain at rest or continue to move in a straight line at constant
velocity unless acted upon by an external force. This is the so-called concept of
inertia.
Law 11
The acceleration of a body is proportional to the resultant force acting on the body,
and this acceleration is in the same direction as the resultant force.
Law III
For every force acting on a body (action) there is an equal and opposite reacting
force (reaction), where the reaction is collinear to the acting force.
3
www.it-ebooks.info
These laws form the basis for much of the analysis in the field of mechanics. Of particular
interest to us in the study of dynamics is the second law, which is written:
F = ma
where F is the resultant force acting on the body, m is the mass of the body, and a is the
linear acceleration of the body’s center of gravity. We’ll discuss this second law in greater
detail later in this chapter, but before that there are some more fundamental issues that
we must address.
Units and Measures
Over years of teaching various engineering courses, we’ve observed that one of the most
common mistakes students make when performing calculations is using the wrong units
for a quantity, thus failing to maintain consistent units and producing some pretty wacky
answers. For example, in the field of ship performance, the most commonly misused
unit is that for speed: people forget to convert speed in knots to speed in meters per
second (m/s) or feet per second (ft/s). One knot is equal to 0.514 m/s, and considering
that many quantities of interest in this field are proportional to speed squared, this
mistake could result in answers that are as much as 185% off target! So, if some of your
results look suspicious later on, the first thing you need to do is go back to your formulas
and check their dimensional consistency.
To check dimensional consistency, you must take a closer look at your units of measure
and consider their component dimensions. We are not talking about 2D or 3D type
dimensions here, but rather the basic measurable dimensions that will make up various
derived units for the physical quantities that we will be using. These basic dimensions
are mass, length, and time.
It is important for you to be aware of these dimensions, as well as the combinations of
these dimensions that make up the other derived units, so that you can ensure dimen¬
sional consistency in your calculations. For example, you know that the weight of an
object is measured in units of force, which can be broken down into component di¬
mensions like so:
F = (M) (L/T 2 )
where M is mass, L is length, and T is time. Does this look familiar? Well, if you consider
that the component units for acceleration are (L/T 2 ) and let a be the symbol for accel¬
eration and m be the symbol for the mass of an object, you get:
F = ma
4 | Chapter 1: Basic Concepts
www.it-ebooks.info
which is the famous expression of Newtons second law of motion. We will take a closer
look at this equation later.
By no means did we just derive this famous formula. What we did was check its di¬
mensional consistency (albeit in reverse), and all that means is that any formulas you
develop to represent a force acting on a body had better come out to a consistent set of
units in the form (M) ( LIT 2 ). This may seem trivial at the moment; however, when you
start looking at more complicated formulas for the forces acting on a body, you’ll want
to be able to break down these formulas into their component dimensions so you can
check their dimensional consistency. Later we will use actual units, from the SI (le Sys-
teme international d’unites, or International System of Units) for our physical quantities.
Of course, there are other unit systems, but unless you want to show these values to
your gamers, it really does not matter which system you use in your games. Again, what
is important is consistency.
To help clarify this point, consider the formula for the friction drag on a body moving
through a fluid, such as water:
R f = 1/2 p V 2 S C f
In this formula, R f represents resistance (a force) due to friction, p is the density of water,
V is the speed of the moving body, S is the submerged surface area of the body, and C f
is an empirical (experimentally determined) drag coefficient for the body. Now rewriting
this formula in terms of basic dimensions instead of variables will show that the di¬
mensions on the left side of the formula match exactly the dimensions on the right side.
Since R f is a force, its basic dimensions are of the form:
(M) (L/T 2 )
as discussed earlier, which implies that the dimensions of all the terms on the right side
of the equation, when combined, must yield an equivalent form. Considering the basic
units for density, speed, and surface area:
• Density: (M)/(L 3 )
. Speed: (L)/(T)
• Area: (L 2 )
and combining these dimensions for the terms, p V 2 S, as follows:
[(M)/(L 3 )] [(L)/(T )] 2 [L 2 ]
and collecting the dimensions in the numerator and denominator yields the following
form:
Units and Measures | 5
www.it-ebooks.info
(M L 2 L 2 ) / (L 3 T 2 )
Canceling dimensions that appear in both the numerator and denominator yields:
M (L/T 2 )
which is consistent with the form shown earlier for resistance, R ( . This exercise also
reveals that the empirical term, C f , for the coefficient of friction must be nondimensional
—that is, it is a constant number with no units.
With that, let’s take a look at some more common physical quantities that you will be
using along with their corresponding symbols, component dimensions, and units in
both the SI and English systems. This information is summarized in Table 1-1.
Table 1-1. Common physical quantities and units
Quantity
Symbol
Dimensions
Units, SI
Units, English
Acceleration, linear
A
L/T 2
m/s 2
ft/s 2
Acceleration, angular
a
radian/T 2
radian/s 2
radian/s 2
Density
P
M/L 3
kg/m 3
slug/ft 3
Force
F
M (L/T 2 )
newton, N
pound, lbs
Kinematic viscosity
V
L 2 /T
m 2 /s
ft 2 /s
Length
L (or x, y, z)
L
meters, m
feet, ft
Mass
m
M
kilogram, kg
slug
Moment (torque)
M a
M (L 2 /T 2 )
N-m
ft-lbs
Mass Moment of Inertia
1
ML 2
kg-m 2
lbs-ft-s 2
Pressure
P
M/(LT 2 )
N/m 2
lbs/ft 2
Time
T
T
seconds, s
seconds, s
Velocity, linear
V
L/T
m/s
ft/s
Velocity, angular
w
radian/T
radian/s
radian/s
Viscosity
P
M/(LT)
Ns/m 2
lbs • s/ft 2
a In general, we will use a capital M to represent a moment (torque) acting on a body and a lowercase m to represent the mass of
a body. If we're referring to the basic dimension of mass in a general sense—that is, referring to the dimensional components of
derived units of measure—well use a capital M. Usually, the meanings of these symbols will be obvious based on the context in
which they are used; however, we will specify their meanings in cases where ambiguity may exist.
Coordinate System
Throughout this book we will refer to a standard, right-handed Cartesian coordinate
system when specifying positions in 2D or 3D space. In two dimensions we will use the
coordinate system shown in Figure 1 -1 (a), where rotations are measured positive coun¬
terclockwise.
6 | Chapter 1: Basic Concepts
www.it-ebooks.info
In three dimensions we will use the coordinate system shown in Figure l-l(b), where
rotations about the x-axis are positive from positive y to positive z, rotations about the
y-axis are positive from positive z to positive x, and rotations about the z-axis are positive
from positive x to positive y.
Vectors
Let us take you back for a moment to your high school math class and review the concept
of vectors. Essentially, a vector is a quantity that has both magnitude as well as direction.
Recall that a scalar, unlike a vector, has only magnitude and no direction. In mechanics,
quantities such as force, velocity, acceleration, and momentum are vectors, and you
must consider both their magnitude and direction. Quantities such as distance, density,
viscosity, and the like are scalars.
With regard to notation, we’ll use boldface type to indicate a vector quantity, such as
force, F. When referring to the magnitude only of a vector quantity, we’ll use standard
type. For example, the magnitude of the vector force, F, is F with components along the
coordinate axes, F x , F , and F z . In the code samples throughout the book, we’ll use the
* (asterisk) to indicate vector dot product, or scalar product, operations depending on
the context, and we’ll use the A (caret) to indicate vector cross product.
Because we will be using vectors throughout this book, it is important that you refresh
your memory on the basic vector operations, such as vector addition, dot product, and
cross product, among others. For your convenience (so you don’t have to drag out that
old math book), we’ve included a summary of the basic vector operations in Appen¬
dix A. This appendix provides code for a Vector class that contains all the important
vector math functionality. Further, we explain how to use specific vector operations—
such as the dot-product and cross-product operations—to perform some common and
Vectors | 7
www.it-ebooks.info
useful, calculations. For example, in dynamics you’ll often have to find a vector per¬
pendicular, or normal, to a plane or contacting surface; you use the cross-product op¬
eration for this task. Another common calculation involves finding the shortest distance
from a point to a plane in space; you use the dot-product operation here. Both of these
tasks are described in Appendix A, which we encourage you to review before delving
too deeply into the example code presented throughout the remainder of this book.
Derivatives and Integrals
If you’re not familiar with calculus, or The Calculus, don’t let the use of derivatives and
integrals in this text worry you. While we’ll write equations using derivatives and inte¬
grals, we’ll show you explicitly how to deal with them computationally throughout this
book. Without going into a dissertation on all the properties and applications of deriv¬
atives and integrals, let’s touch on their physical significance as they relate to the material
we’ll cover.
You can think of a derivative as the rate of change in one variable with respect to another
variable, or in other words, derivatives tells you how fast one variable changes as some
other variable changes. Take speed, for example. A car travels at a certain speed covering
some distance in a certain period of time. Its speed, on average, is the distance traveled
over a specific time interval. If it travels a distance of 60 kilometers in one hour, then
its average speed is 60 kilometers an hour. When we’re doing simulations, the ones you’ll
see later in this book, we’re interested in what the car is doing over very short time
intervals. As the time interval gets really small and we consider the distance traveled
over that very short period of time, we’re looking at instantaneous speed. We usually
write such relations using symbols like the following:
Ivl = ds/dt
where v is the speed, ds is a small distance (a differential distance), and dt is a small,
differential, period of time. In reality, for our simulations, we’ll never deal with infinitely
small numbers; we’ll use small numbers, such as time intervals of 1 millisecond, but not
infinitely small numbers.
For our purposes, you can think of integrals as the reverse, or the inverse, of derivatives;
integration is the inverse of differentiation. The symbol J represents integration. You
can think of integration as a process of adding up a bunch of infinitely small chunks of
some variable. Here again, we are not going to deal with infinitely small pieces of any¬
thing, but instead will consider small, discrete parcels of some variable—for example,
a small, discrete amount of time, area, or mass. In these cases, we’ll use the Z symbol
instead of the integration symbol. Consider a loaf of bread that’s sliced into uniformly
thick slices along its whole length. If you wanted to compute the volume of that loaf of
bread, you can approximate it by starting at one end and computing the volume of the
first slice, approximating its volume as though it were a very short, square cylinder; then
8 | Chapter 1: Basic Concepts
www.it-ebooks.info
moving on to the second slice, estimating its volume and adding that to the volume of
the first slice; and then moving on to the third, and fourth, and so on, aggregating the
volume of the loaf as you move toward the other end. Integration applies this technique
to infinitely thin slices of volume to compute the volume of any arbitrary shape. The
same techniques apply to other computations—for example, computing areas, iner¬
tias , masses, and so on, and even aggregating distance traveled over successive small
slices of time, as you’ll see later. In fact, this latter application is the inverse of the de¬
rivative of distance with respect to time, which gives speed. Using integration and dif¬
ferentiation in this way allows you to work back and forth when computing speed,
acceleration, and distance traveled, as you’ll see shortly. In fact, we’ll use these concepts
heavily throughout the rest of this book.
Mass, Center of Mass, and Moment of Inertia
The properties of a body— mass, center of mass, and moment of inertia, collectively called
mass properties —are absolutely crucial to the study of mechanics, as the linear and
angular 1 motion of a body and a body’s response to a given force are functions of these
mass properties. Thus, in order to accurately model a body in motion, you need to know
or be capable of calculating these mass properties. Let’s look at a few definitions first.
In general, people think of mass as a measure of the amount of matter in a body. For
our purposes in the study of mechanics, we can also think of mass as a measure of a
body’s resistance to motion or a change in its motion. Thus, the greater a body’s mass,
the harder it will be to set it in motion or change its motion.
In laymen’s terms, the center of mass (also known as center of gravity) is the point in a
body around which the mass of the body is evenly distributed. In mechanics, the center
of mass is the point through which any force can act on the body without resulting in
a rotation of the body.
Although most people are familiar with the terms mass and center of gravity, the term
moment of inertia is not so familiar; however, in mechanics it is equally important. The
mass moment of inertia of a body is a quantitative measure of the radial distribution of
the mass of a body about a given axis of rotation. Analogous to mass being a measure
of a body’s resistance to linear motion, mass moment of inertia (also known as rotational
inertia) is a measure of a body’s resistance to rotational motion.
Now that you know what these properties mean, let’s look at how to calculate each.
For a given body made up of a number of particles, the total mass of the body is simply
the sum of the masses of all elemental particles making up the body, where the mass of
1. Linear motion refers to motion in space without regard to rotation; angular motion refers specifically to the
rotation of a body about any axis (the body may or may not be undergoing linear motion at the same time).
Mass, Center of Mass, and Moment of Inertia | 9
www.it-ebooks.info
each elemental particle is its mass density times its volume. Assuming that the body is
of uniform density, then the total mass of the body is simply the density of the body
times the total volume of the body. This is expressed in the following equation:
m = /pdV = p f dV
In practice, you rarely need to take the volume integral to find the mass of a body,
especially considering that many of the bodies we will consider—for example, cars and
planes—are not of uniform density. Thus, you will simplify these complicated bodies
by breaking them down into an ensemble of component bodies of known or easily
calculable mass and simply sum the masses of all components to arrive at the total mass.
The calculation of the center of gravity of a body is a little more involved. First, divide
the body into a finite number of elemental masses with the center of each mass specified
relative to the reference coordinate system axes. Well refer to these elemental masses
as m r Next, take the first moment of each mass about the reference axes and then add
up all of these moments. The first moment is the product of the mass times the distance
along a given coordinate axis from the origin to the center of mass. Finally, divide this
sum of moments by the total mass of the body, yielding the coordinates to the center of
mass of the body relative to the reference axes. You must perform this calculation once
for each dimension—that is, twice when working in 2D and three times when working
in 3D. Here are the equations for the 3D coordinates of the center of mass of a body:
x c = { fx 0 dm} / m
Yc= {/y 0 dm} /m
z c = {/ z o dm} /m
where (x, y, z) c are the coordinates of the center of mass for the body and (x, y, z) 0 are
the coordinates of the center of mass of each elemental mass. The quantities x a dm, y a
dm, and z a dm represent the first moments of the elemental mass, dm, about each of the
coordinate axes.
Here again, don’t worry too much about the integrals in these equations. In practice,
you will be summing finite numbers of masses and the formulas will take on the friend¬
lier forms shown here:
x c = {E x 0 mj} / {E m;}
y c = {E y 0 mj} / {Em;}
z c = {E z 0 nii} / {Emi}
Note that you can easily substitute weights for masses in these formulas since the con¬
stant acceleration due to gravity, g, would appear in both the numerators and denomi-
10 | Chapter 1: Basic Concepts
www.it-ebooks.info
nators, thus dropping out of the equations. Recall that the weight of an object is its mass
times the acceleration due to gravity, g, which is 9.8 m/s 2 at sea level.
The formulas for calculating the total mass and center of gravity for a system of discrete
point masses can conveniently be written in vector notation as follows:
m t = X mj
CG=[S (c gi ) ( mi )]/m t
where m, is the total mass, m i is the mass of each point mass in the system, CG is the
combined center of gravity, and cgj is the location of the center of gravity of each point
mass in design, or reference, coordinates. Notice that CG and eg, are shown as vectors
since they denote position in Cartesian coordinates. This is a matter of convenience
since it allows you to take care of the x, y, and z components (or just x and y in two
dimensions) in one shot.
In the code samples that follow, let’s assume that the point masses making up the body
are represented by an array of structures where each structure contains the point mass’s
design coordinates and mass. The structure will also contain an element to hold the
coordinates of the point mass relative to the combined center of gravity of the rigid
body, which will be calculated later.
typedef struct _PointMass
{
float mass;
Vector designPosition;
Vector correctedPosition;
} PointMass;
// Assume that _NUMELEMENTS has been defined
PointMassElements[_NUMELEMENTS];
Here’s some code that illustrates how to calculate the total mass and combined center
of gravity of the elements:
int i;
float TotalMass;
Vector CombinedCG;
Vector FlrstMoment;
TotalMass = 0;
for(i=0; i<_NUMELEMENTS; i++)
TotalMass FAC+= Elements[i].mass;
FlrstMoment = Vector(0, 0, 0);
for(l=0; i<_NUMELEMENTS; i++)
{
FlrstMoment += Element[i].mass * Element[i].designPosttion;
Mass,CenterofMass,andMomentoflnertia | 11
www.it-ebooks.info
}
ConblnedCG = FlrstMonent / TotalMass;
Now that the combined center of gravity location has been found, you can calculate the
relative position of each point mass as follows:
for(t=0; i<_NUMELEMENTS; i++)
{
Elenentfi].correctedPosition = Elenentfi].designPosition -
ComblnedCG;
}
To calculate mass moment of inertia, you need to take the second moment of each
elemental mass making up the body about each coordinate axis. The second moment
is then the product of the mass times distance squared. That distance is not the distance
to the elemental mass centroid along the coordinate axis as in the calculation for center
of mass, but rather the perpendicular distance from the coordinate axis, about which
we want to calculate the moment of inertia, to the elemental mass centroid.
Referring to Figure 1-2 for an arbitrary body in three dimensions, when calculating
moment of inertia about the x-axis, / xx , this distance, r, will be in the yz-plane such that
r 2 = y 2 + z 2 . Similarly, for the moment of inertia about the y-axis, / yy , r 2 = z 2 + x 2 , and
for the moment of inertia about the z-axis, / zz , r 2 = x 2 + y 2 .
Figure 1-2. Arbitrary body in 3D
The equations for mass moment of inertia about the coordinate axes in 3D are:
Ixx = /r x 2 dm = /(y 2 + z 2 ) dm
lyy = fr y 2 dm = /(z 2 + x 2 ) dm
Izz = /r z 2 dm = / (x 2 + y 2 ) dm
Let’s look for a moment at a common situation that arises in practice. Say you are given
the moment of inertia, I a , of a body about an axis, called the neutral axis, passing through
12 | Chapter 1: Basic Concepts
www.it-ebooks.info
the center of mass of the body, but you want to know the moment of inertia, I, about an
axis some distance from but parallel to this neutral axis. In this case, you can use the
transfer of axes, or parallel axis theorem, to determine the moment of inertia about this
new axis. The formula to use is:
T = I 0 + md 2
where m is the mass of the body and d is the perpendicular distance between the parallel
axes.
There is an important practical observation to make here: the new moment of inertia
is a function of the distance separating the axes squared. This means that in cases where
I a is known to be relatively small and d relatively large, you can safely ignore /„, since
the md 2 term will dominate. You must use your best judgment here, of course. This
formula for transfer of axes also indicates that the moment of inertia of a body will be
at its minimum when calculated about an axis passing through the body’s center of
gravity. The body’s moment of inertia about any parallel axis will always increase by an
amount, md 2 , when calculated about an axis not passing through the body’s center of
mass.
In practice, calculating mass moment of inertia for all but the simplest shapes of uniform
density is a complicated endeavor, so we will often approximate the moment of inertia
of a body about axes passing through its center of mass by using simple formulas for
basic shapes that approximate the object. Further, we will break down complicated
bodies into smaller components and take advantage of the fact that I a may be negligible
for certain components considering its md 2 contribution to the total body’s moment of
inertia.
Figure 1-3 through Figure 1-7 show some simple solid geometries for which you can
easily calculate mass moments of inertia. The mass moment of inertia formulas for each
of these simple geometries of homogenous density about the three coordinate axes are
shown in the figure captions. You can readily find similar formulas for other basic ge¬
ometries in college-level dynamics texts (see the Bibliography at the end of this book
for a few sources).
Mass, Center of Mass, and Moment of Inertia | 13
www.it-ebooks.info
Figure 1-5. Rectangular cylinder: I xx = (1/12) m(a 2 + l 2 ); l yy = (1/12) m(b 2 + l 2 ); I z: =
(1/12) m(a 2 + b 2 )
14 | Chapter 1: Basic Concepts
www.it-ebooks.info
;
<
Figure 1-6. Sphere: 1^ = I yy = I zz = (2/5) mr
;
Z'^^'
..■V Xy
Figure 1-7. Spherical shell: I xx = I = I zz = (2/3) mr 2
As you can see, these formulas are relatively simple to implement. The trick here is to
break up a complex body into a number of smaller, simpler representative geometries
whose combination will approximate the complex body’s inertia properties. This exer¬
cise is largely a matter of judgment considering the desired level of accuracy.
Let’s look at a simple 2D example demonstrating how to apply the formulas discussed
in this section. Suppose you’re working on a top-down-view auto racing game where
you want to simulate the automobile sprite based on 2D rigid-body dynamics. At the
start of the game, the player’s car is at the starting line, full of fuel and ready to go. Before
starting the simulation, you need to calculate the mass properties of the car, driver, and
fuel load at this initial state. In this case, the body is made up of three components: the
car, driver, and full load of fuel. Later during the game, however, the mass of this body
will change as fuel burns off and the driver gets thrown after a crash! For now, let’s focus
on the initial condition, as illustrated in Figure 1-8.
Mass,CenterofMass,andMomentoflnertia | 15
www.it-ebooks.info
The properties of each component in this example are given in Table 1-2. Note that
length is measured along the x-axis, width along the y-axis, and height would be coming
out of the screen. Also note that the coordinates—in the form (x,y )—to the centroid of
each component are referenced to the global origin.
Table 1-2. Example properties
1 Car
Driver (seated)
Fiicl 1
Length = 4.70 m
Length = 0.90 m
Length = 0.50 m
Width = 1.80 m
Width = 0.50 m
Width = 0.90 m
Height = 1.25 m
Height= 1.10 m
Height = 0.30 m
Weight = 17,500 N
Weight = 850 N
Density of fuel = 750 kg/m 3
Centroid = (30.5,30.5) m
Centroid = (31.50,31.00) m
Centroid = (28.00,30.50) m
The first mass property we want to calculate is the mass of the body. This is a simple
calculation since we are already given the weight of the car and the driver. The only
other component of weight we need is that of the fuel. Since we are given the mass
density of the fuel and the geometry of the tank, we can calculate the volume of the tank
and multiply by the density and the acceleration due to gravity to get the weight of the
fuel in the tank. This yields 920.6 N of fuel, as shown here:
Wfuei = pvg = (750 kg/m 3 ) (0.50 m) (0.90 m) (0.30 m) (9.81 m/s 2 )
= 993 N
-
0 %
•>' 4 »,
a ,4 *
Acceleration due to gravity is the acceleration of a falling object as it
falls toward the earth. The weight of an object is equal to its mass times
the acceleration due to gravity. The symbol g is used to represent the
acceleration due to gravity, and on Earth the value ofgis approximately
9.8 m/s 2 at sea level. Units for weight in the metric system are Newtons,
N.
16 | Chapter 1: Basic Concepts
www.it-ebooks.info
Now, the total weight of the body is:
Wtotal = W car + Wjj-j ver + Wf ue i
Wtotal = 17,500 N + 850 N + 993 N = 19,343 N
To get the mass of the body, you simply divide the weight by the acceleration due to
gravity.
Mtotal = Wt 0 tai/g = 19,343 N /(9.81 m/s 2 ) = 1972 kg
The next mass property we want is the location of the center of gravity of the body. In
this example we will calculate the centroid relative to the global origin. We will also apply
the first moment formula twice, once for the x coordinate and again for they coordinate:
X C g body = {( x cg car)(W ca r) + (x C g driverX^driver) + (x C g fuel)
(Wfuel)} / Wtotal
X C g body = {(30.50 m)( 17,500 N) + (31.50 m)(850 N) + (28.00 m)
(993 N)} / 19,343 N
Xcg body = 30.42 m
Ycg body = {(y C g car)(W ca r) + (Ycg driver)(^driver) + (Ycg fuel)
(Wfuei)} /Wt 0 tal
Y cg body = {(30.50 m)(17,500 N) + (31.00 m)(850 N) + (30.50 m)
(993 N)} / 19,343 N
Ycg body = 30.52 m
Notice that we used weight in these equations instead of mass. Remember we can do
this because the acceleration due to gravity built into the weight value is constant and
appears in both the numerator and denominator, thus canceling out.
Now it’s time to calculate the mass moment of inertia of the body. This is easy enough
in this 2D example since we have only one rotational axis, coming out of the paper, and
thus need only perform the calculation once. The first step is to calculate the local
moment of inertia of each component about its own neutral axis. Given the limited
information we have on the geometry and mass distribution of each component, we
will make a simplifying approximation by assuming that each component can be rep¬
resented by a rectangular cylinder, and will thus use the corresponding formula for
moment of inertia from Figure 1-5. In the equations to follow, we’ll use a lowercase w
to represent width so as to not confuse it with weight, where we’ve been using a capital
W.
I 0 car = (m/12) (W 2 + L 2 )
Mass,CenterofMass,andMomentoflnertia | 17
www.it-ebooks.info
Io car = ((17,500 N / 9.81 m/s 2 ) / 12 ) ((1.80 m) 2 + (4.70 m) 2 ) =
3765.5 N - s 2 - m
I 0 driver = (m/12) (w 2 + L 2 )
Io driver = ((850 N / 9.81 m/s 2 ) / 12) ((0.50 m) 2 + (0.90 m) 2 ) = 7.7
N - s 2 - m
Io fuel = (m/12) (w 2 + L 2 )
Io fuel = ((993 N / 9.81 m/s 2 ) / 12) ((0.90 m) 2 + (0.50 m) 2 ) = 8.9 N
- s 2 - m
Since these are the moments of inertia of each component about its own neutral axis,
we now need to use the parallel axis theorem to transfer these moments to the neutral
axis of the body, which is located at the body center of gravity that we recently calculated.
To do this, we must find the distance from the body center of gravity to each component’s
center of gravity. The distances squared from each component to the body center of
gravity are:
d“car — ( x cg car X C g) _ + (Ycg car Y C g) _
d 2 car = (30.50 m - 30.42 m) 2 + (30.50 m - 30.53 m) 2 = 0.01 m 2
)“ + (y C g driver Y C g) 2
d 2 driver = (31.50 m - 30.42 m) 2 + (31.25 m - 30.53 m) 2 = 1.68
^“driver — ( x cg driver X C g)“ + (ycg driver ^cg)"
m-
2 fuel =
( x cg fuel X cg )- + (ycg fuel 5f C g)
d 2 f U ei = (28.00 m - 30.42 m) 2 + (30.50 m - 30.53 m) 2 = 5.86 m 2
Now we can apply the parallel axis theorem as follows:
leg car = Io + md“
leg car = 3765.5 N - s 2 - m + (17,500 N / 9.81 m/s 2 ) (0.01 m 2 ) =
3783.34 N-s 2 -m
leg driver = Io + md -
Icg driver = 7.7 N - s 2 - m + (850 N / 9.81 m/s 2 ) (1.68 m 2 ) =
153.27 N - s 2 - m
leg fuel = Io + md“
leg fuel = 8.9 N - s 2 - m + (993 N / 9.81 m/s 2 ) (5.86 m 2 ) =
602.07 N - s 2 - m
18 | Chapter 1: Basic Concepts
www.it-ebooks.info
Notice how the calculations for the I cg of the driver and the fuel are dominated by their
md 2 terms. In this example, the local inertia of the driver and fuel is only 2.7% and 2.1%,
respectively, of their corresponding md 2 terms.
Finally, we can obtain the total moment of inertia of the body about its own neutral axis
by summing the I cg contributions of each component as follows:
leg total = leg car + leg driver + ^cg fuel
leg total = 3783.34 N - s 2 - m + 153.27 N - s 2 - m + 602.07 N -
s 2 - m = 4538.68 N - s 2 - m
The mass properties of the body—that is, the combination of the car, driver, and full
tank of fuel—are shown in Table 1-3.
Table 1-3. Example summary of mass properties
Property Computed value
Total mass (weight) 1972 kg (19,343 N)
Combined center of mass location (x,y) = (30.42 m, 30.53 m)
Mass moment of inertia 4538.68 N - s 2 - m
It is important that you understand the concepts illustrated in this example well because
as we move on to more complicated systems and especially to general motion in 3D,
these calculations are only going to get more complicated. Moreover, the motion of the
bodies to be simulated are functions of these mass properties, where mass will determine
how these bodies are affected by forces, center of mass will be used to track position,
and mass moment of inertia will determine how these bodies rotate under the action of
noncentroidal forces.
So far, we have looked at moments of inertia about the three coordinate axes in 3D space.
However, in general 3D rigid-body dynamics, the body may rotate about any axis—not
necessarily one of the coordinate axes, even if the local coordinate axes pass through
the body center of mass. This complication implies that we must add a few more terms
to our set of Is for a body to handle this generalized rotation. We will address this topic
further later in this chapter, but before we do that we need to go over Newton’s second
law of motion in detail.
Mass,CenterofMass,andMomentoflnertia | 19
www.it-ebooks.info
Newton's Second Law of Motion
As we stated in the first section of this chapter, Newton’s second law of motion is of
particular interest in the study of mechanics. Recall that the equation form of Newtons
second law is:
F = ma
where F is the resultant force acting on the body, m is the mass of the body, and a is the
linear acceleration of the body center of gravity.
If you rearrange this equation as follows:
F/m = a
you can see how the mass of a body acts as a measure of resistance to motion. Observe
here that as mass increases in the denominator for a constant applied force, then the
resulting acceleration of the body will decrease. You could say that the body of greater
mass offers greater resistance to motion. Similarly, as the mass decreases for a constant
applied force, then the resulting acceleration of the body will increase, and you could
say that the body of smaller mass offers lower resistance to motion.
Newton’s second law also states that the resulting acceleration is in the same direction
as the resultant force on the body; thus, force and acceleration must be treated as vector
quantities. In general, there may be more than one force acting on the body at a given
time, which means that the resultant force is the vector sum of all forces acting on the
body. So, you can now write:
Yj F = ma
where a represents the acceleration vector.
In 3D, the force and acceleration vectors will have x,y, and z components in the Cartesian
reference system. In this case, the component equations of motion are written as follows:
Z F x = ma x
Z Fy = may
ZF z = ma z
An alternative way to interpret Newton’s second law is that the sum of all forces acting
on a body is equal to the rate of change of the body’s momentum over time, which is
the derivative of momentum with respect to time. Momentum equals mass times ve¬
locity, and since velocity is a vector quantity, so is momentum. Thus:
20 | Chapter 1: Basic Concepts
www.it-ebooks.info
G = mv
where G is linear momentum of the body, m is the body’s mass, and v is velocity of the
center of gravity of the body. The time rate of change of momentum is the derivative of
momentum with respect to time:
dG/dt = d/dt (mv)
Assuming that the body mass is constant (for now), you can write:
dG/dt = m dv/dt
Observing that the time rate of change of velocity, dv/dt, is acceleration, we arrive at:
dG/dt = ma
and:
Yj F = dG/dt = ma
So far we have considered only translation of the body without rotation. In generalized
3D motion, you must account for the rotational motion of the body and will thus need
some additional equations to fully describe the body’s motion. Specifically, you will
require analogous formulas relating the sum of all moments (torque) on a body to the
rate of change in its angular momentum over time, or the derivative of angular mo¬
mentum with respect to time. This gives us:
I M cg = d/dt (H cg )
where X M cg is the sum of all moments about the body center of gravity, and H is the
angular momentum of the body. M cg can be expressed as:
M cg = r x F
where F is a force acting on the body, and r is the distance vector from F, perpendicular
to the line of action of F (i.e., perpendicular to the vector F), to the center of gravity of
the body, and x is the vector cross-product operator.
The angular momentum of the body is the sum of the moments of the momentum of
all particles in the body about the axis of rotation, which in this case we assume passes
through the center of gravity of the body. This can be expressed as:
H cg = 2 rj x mj (to x rj)
Newton's Second Law of Motion | 21
www.it-ebooks.info
where i represents the zth particle making up the body, to is the angular velocity of the
body about the axis under consideration, and (r ; x m { (to x r,)) is the angular momentum
of the zth particle, which has a magnitude of m^r 2 . For rotation about a given axis, this
equation can be rewritten in the form:
H cg = /tor dm
Given that the angular velocity is the same for all particles making up the rigid body,
we have:
H cg = to Jr 2 dm
and recalling that moment of inertia, I, equals / r 2 dm, we get:
H cg = Ito
Taking the derivative with respect to time, we obtain:
dH cg /dt = d/dt (Ito) = I dto/dt = la
where a is the angular acceleration of the body about a given axis.
Finally, we can write:
S M cg = la
As we stated in our discussion on mass moment of inertia, we will have to further
generalize our formulas for moment of inertia and angular moment to account for
rotation about any body axis. Generally, M and a will be vector quantities, while I will
be a tensor 2 since the magnitude of moment of inertia for a body may vary depending
on the axis of rotation (see the sidebar “Tensors” on page 22).
Tensors
A tensor is a mathematical expression that has magnitude and direction, but its mag¬
nitude may not be unique depending on the direction. Tensors are typically used to
represent properties of materials where these properties have different magnitudes in
different directions. Materials with properties that vary depending on direction are
called anisotropic (isotropic implies the same magnitude in all directions). For example,
consider the elasticity (or strength) of two common materials, a sheet of plain paper
2. In this case, I will be a second-rank tensor, which is essentially a 3x3 matrix. A vector is actually a tensor of
rank one, and a scalar is actually a tensor of rank zero.
22 | Chapter 1: Basic Concepts
www.it-ebooks.info
and a piece of woven or knitted cloth. Take the sheet of paper and, holding it flat, pull
on it softly from opposing ends. Try this length-wise, width-wise, and along a diagonal.
You should observe that the paper seems just as strong, or stretches about the same, in
all directions. It is isotropic; therefore, only a single scalar constant is required to rep¬
resent its strength for all directions.
Now, try to find a piece of cloth with a simple, relatively loose weave where the threads
in one direction are perpendicular to the threads in the other direction. Most neckties
will do. Try the same pull test that you conducted with the sheet of paper, pulling the
cloth along each thread direction and then at a diagonal to the threads. You should
observe that the cloth stretches more when you pull it along a diagonal to the threads
as opposed to pulling it along the direction of the run of the threads. The cloth is ani¬
sotropic in that it exhibits different elastic (or strength) properties depending on the
direction of pull; thus, a collection of vector quantities (a tensor) is required to represent
its strength for all directions.
In the context of this book, the property under consideration is a body’s moment of
inertia, which in 3D requires nine components to fully describe it for any arbitrary
rotation. Moment of inertia is not a strength property as in the paper and cloth example,
but it is a property of the body that varies with the axis of rotation. Since nine components
are required, moment of inertia will be generalized in the form of a 3x3 matrix (i.e., a
second-rank tensor) later in this book.
We need to mention a few things at this point regarding coordinates, which will become
important when you’re writing your real-time simulator. Both of the equations of mo¬
tion have, so far, been written in terms of global coordinates and not body-fixed coor¬
dinates. That’s OK for the linear equation of motion, where you can track the body’s
location and velocity in the global coordinate system. However, from a computational
point of view, you don’t want to do that for the angular equation of motion for bodies
that rotate in three dimensions. 3 The reason is because the moment of inertia term,
when calculated with respect to global coordinates, actually changes depending on the
body’s position and orientation. This means that during your simulation you’ll have to
recalculate the inertia matrix (and its inverse) a lot, which is computationally inefficient.
It’s better to rewrite the equations of motion in terms of local (attached to the body)
coordinates so you have to calculate the inertia matrix (and its inverse) only once at the
start of your simulation.
In general, the time derivative of a vector, V, in a fixed (nonrotating) coordinate system
is related to its time derivative in a rotating coordinate system by the following:
(dV/dt) fixed = (dV/dt) rot + (® X V)
3. In two dimensions, its OK to leave the angular equation of motion as its shown here since the moment of
inertia term is simply a constant scalar quantity.
Newton's Second Law of Motion | 23
www.it-ebooks.info
The (to x V) term represents the difference between Vs time derivative as measured in
the fixed coordinate system and V’s time derivative as measured in the rotating coor¬
dinate system. We can use this relation to rewrite the angular equation of motion in
terms of local, or body-fixed, coordinates. Further, the vector to consider is the angular
momentum vector H cg . Recall that H cg = I to and its time derivative are equal to the sum
of moments about the body’s center of gravity. These are the pieces you need for the
angular equation of motion, and you can get to that equation by substituting H cg in place
of V in the derivative transform relation as follows:
£ M cg = dH cg /dt = I (doo/dt) + (<d X (I ©))
where the moments, inertia tensor, and angular velocity are all expressed in local (body)
coordinates. Although this equation looks a bit more complicated than the one we
showed you earlier, it is much more convenient to use since I will be constant throughout
your simulation (unless your body’s mass or geometry changes for some reason during
your simulation), and the moments are relatively easy to calculate in local coordinates.
You’ll put this equation to use in Chapter 15 when we show you how to develop a simple
3D rigid-body simulator.
Inertia Tensor
Take another look at the angular equation of motion and notice that we set the inertia
term, I, in bold, implying that it is a vector. You’ve already seen that, for two-dimensional
problems, this inertia term reduces to a scalar quantity representing the moment of
inertia about the single axis of rotation. However, in three dimensions there are three
coordinate axes about which the body can rotate. Moreover, in generalized three di¬
mensions, the body can rotate about any arbitrary axis. Thus, for three-dimensional
problems, I is actually a 3x3 matrix—a second-rank tensor.
To understand where this inertia matrix comes from, you must look again at the angular
momentum equation:
H cg = /(r x (<d x r)) dm
where to is the angular velocity of the body; r is the distance from the body’s center of
gravity to each elemental mass, dm; and (r x (to x r ))dm is the angular momentum of
each elemental mass. The term in parentheses is called a triple vector product and can
be expanded by taking the vector cross products, r and to are vectors that can be written
as follows:
r = xi + yj + zk
(O = (% i + (Dy j + co z k
24 [ Chapter 1: Basic Concepts
www.it-ebooks.info
Expanding the triple vector product term yields:
H cg = / { [(y 2 + z 2 )o) x - xyo) y - xzm z ] i +
[ -yxw x + (z 2 + x 2 )(o y - yzo) z ] j +
[-zxo) x - zytOy + (x 2 + y 2 )coz] k} dm
To simplify this equation, let’s replace a few terms by letting:
Ixx = / (y 2 + z 2 ) dm
lyy = /(Z 2 + X 2 ) dm
Izz = /(x 2 + y 2 ) dm
Ixy = Jyx = /( x Y) dm
I xz — Izx = /( xz ) dm
lyz = Izy = / (yz) dm
Substituting these I variables, some of which should look familiar to you, back into the
expanded equation yields:
H C g = [I xx <«>x — Ixy ®y — Ixz ®z] I +
[— Iyx w x + lyy ©y — lyz (l, zl j +
L — Izx ©X — Izy ©y + Izz ©zl ^
Simplifying this a step further by letting I be a matrix:
Ixx — Ixy — Ixz
I = — lyx lyy — lyz
— Izx — Izy Izz
yields the following equation:
H cg = I ©
Inertia Tensor | 25
www.it-ebooks.info
You already know that I represents the moment of inertia, and the terms that should
look familiar to you already are the moment of inertia terms about the three coordinate
axes, 1xx, Iyy, and f zz . The other terms are called products of inertia (see Figure 1-9):
Ixy = Iy x = / (xy) dm
I xz — Izx = /( xz ) dm
lyz = Izy = / (yz) dm
Figure 1-9. Products of inertia
Just like the parallel axis theorem, there’s a similar transfer of axis formula that applies
to products of inertia:
Ixy = lo(xy) + m d x dy
Ixz = lo(xz) + m d x d z
lyz = lo(yz) + m dy d z
where the I 0 terms represent the local products of inertia (that is, the products of inertia
of the object about axes that pass through its own center of gravity), m is the object’s
mass, and the d terms are the distances between the coordinate axes that pass through
the object’s center of gravity and a parallel set of axes some distance away (see
Figure 1-10).
You’ll notice that we did not give you any product of inertia formulas for the simple
shapes shown earlier in Figure 1-3 through Figure 1-7. The reason is that the given
moments of inertia were about the principal axes for these shapes. For any body, there
exists a set of axes—called the principal axes—oriented such that the product of inertia
terms in the inertia tensor are all zero.
26 | Chapter 1: Basic Concepts
www.it-ebooks.info
y 0
Figure 1-10. Transfer of axes
For the simple geometries shown earlier, each coordinate axis represented a plane of
symmetry, and products of inertia go to zero about axes that represent planes of sym¬
metry. You can see this by examining the product of inertia formulas, where, for ex¬
ample, all of the (xy) terms in the integral will be cancelled out by each corresponding
-(xy) term if the body is symmetric about the y-axis, as illustrated in Figure 1-11.
-
/
dm -x
X
dm
y
y
Figure 1-11. Symmetry
For composite bodies, however, there may not be any planes of symmetry, and the
orientation of the principal axes will not be obvious. Further, you may not even want to
use the principal axes as your local coordinate axes for a given rigid body since it may
Inertia Tensor | 27
www.it-ebooks.info
be awkward to do so. For example, consider the airplane from the FlightSim discussion
in Chapter 7, where you’ll have the local coordinate design axes running, relative to the
pilot, fore and aft, up and down, and left and right. This orientation is convenient for
locating the parts of the wings, tail, elevators, etc. with respect to one another, but these
axes don’t necessarily represent the principal axes of the airplane. The end result is that
you’ll use axes that are convenient and deal with the nonzero products of inertia (which,
by the way, can be either positive or negative).
We already showed you how to calculate the combined moments of inertia for a com¬
posite body made up of a few smaller elements. Accounting for the product of inertia
terms follows the same procedure except that, typically, your elements are such that
their local product of inertia terms are zero. This is the case only if you represent your
elements by simple geometries such as point masses, spheres, rectangles, etc. That being
the case, the main contribution to the rigid body’s products of inertia will be due to the
transfer of axes terms for each element.
Before looking at some sample code, let’s first revise the element structure to include a
new term to hold the element’s local moment of inertia as follows:
typedef struct _PointMass
1
float mass;
Vector designPositlon;
Vector correctedPositlon;
Vector locallnertla;
} PointMass;
Here we’re using a vector to represent the three local moment of inertia terms and we’re
also assuming that the local products of inertia are zero for each element.
The following code sample shows how to calculate the inertia tensor given the compo¬
nent elements:
float Ixx, Iyy, Izz, Ixy, Ixz, Iyz;
Matrix3x3 InertiaTensor;
Ixx = 0; Iyy = 0; Izz = 0;
Ixy = 0; Ixz = 0; Iyz = 0;
for (i = 0; i<_NUMELEMENTS; 1++)
1
Ixx += Elenentfl].Locallnertla.x +
Elenentfl].mass * (Elementfi].correctedPositlon.y *
Elementfl].correctedPositlon.y +
Elementfi].correctedPositlon.z *
Element[i].correctedPositlon.z);
Iyy += Elementfl].Locallnertla.y +
Elementfi].mass * (Elementfi].correctedPositlon.z *
Elementfi].correctedPositlon.z +
28 | Chapter 1: Basic Concepts
www.it-ebooks.info
Elemental] .correctedPosition.x *
Element[i].correctedPosition.x);
Izz += Element[i].Locallnertia.z +
Elementfl].mass * (Element[i].correctedPosition.x *
Element[i].correctedPosition.x +
Element[i].correctedPosition.y *
Element[i].correctedPosition.y);
Ixy += Element[i].mass * (Element[i].correctedPosition.x *
Element[i].correctedPosition.y);
Ixz += Element[i].mass * (Element[i].correctedPosition.x *
Element[i].correctedPosition.z);
Iyz += Element[i].mass * (Element[i].correctedPosition.y *
Element[i].correctedPosition.z);
}
// ell stands for element on row 1 column 1, el2 for row 1 column 2, etc.
InertiaTensor.ell = Ixx;
InertiaTensor,el2 = -Ixy;
InertiaTensor.el3 = -Ixz;
InertiaTensor,e21 = -Ixy;
InertiaTensor,e22 = Iyy;
InertiaTensor.e23 = -Iyz;
InertiaTensor,e31 = -Ixz;
InertiaTensor,e32 = -Iyz;
InertiaTensor.e33 = Izz;
Note that the inertia tensor is calculated about axes that pass through the combined
center of gravity for the rigid body, so be sure to use the corrected coordinates for each
element relative to the combined center of gravity when applying the transfer of axes
formulas.
We should also mention that this calculation is for the inertia tensor in body-fixed
coordinates, or local coordinates. As we discussed earlier in this chapter, it is better to
rewrite the angular equation of motion in terms of local coordinates and use the local
inertia tensor to save some number crunching in your real-time simulation.
Relativistic Time
To allow for a thorough understanding of how advanced space vehicles work as well as
give you a mechanism by which to alter time in your games, we would like to offer a
brief introduction to the theory of relativity, and particularly its effect on time. In our
everyday experience, it is safe to assume that the clock on your wall is ticking at the same
rate as the clock on our wall as we write this. However, the reason we all know the name
Relativistic Time | 29
www.it-ebooks.info
Albert Einstein is that he had the foresight to abandon time as a constant. Instead he
postulated that light travels at the same speed regardless of the motion of the source.
That is to say, if you shine a flashlight in a vacuum, the electromagnetic radiation it emits
in the form of visible light travels at a set velocity of c (299,792,458 m/s). Now, if you
take that same flashlight and put it on the nose of a rocket traveling at half that speed
directly at you, you might expect that light is traveling at you with a velocity of 1.5c. Yet,
the rocket-powered flashlight would still be observed as emitting light at a velocity of
c. As Einstein’s theory of special relativity matured, the postulate has been reformulated
to state that there is a maximum speed at which information can be transferred in the
space-time continuum, a principal called locality. As electromagnetic radiation has no
mass, 4 it travels at this maximum speed in a vacuum.
The most startling consequence of the theory is that time is no longer absolute. The
postulate that the speed of light is constant for all frames of reference requires that time
slow down, or dilate, as velocity increases. It is actually fairly easy to demonstrate this
result.
The following example depicts a conceptual clock. A beam of light is bouncing between
two mirrors. The time it takes for the beam of light to start from one mirror, bounce off
the second, and return to the first constitutes one “tick” of this clock. That tick can be
calculated as:
At = 2L/c
Where L is the distance between the mirrors and c is the speed of light. Figure 1-12
shows what the clock would look like if you were above it traveling at its velocity.
Mirror A
-—-r
" A
L
___ ±.
Mirror B
Figure 1-12. Traveling with the clock
4. Photons, the particle form of electromagnetic radiation, can have relativistic mass but are hypothesized to
have no “rest mass.” To avoid getting into quantum electrodynamics, here we’ll just consider them without
mass.
30 | Chapter 1: Basic Concepts
www.it-ebooks.info
Now suppose that you are above the mirrors as they speed past you to the right. Then
the clock would look something like Figure 1-13.
One tick of the clock is now defined as twice the distance of the hypotenuse over the
speed of light. Clearly H must be larger than L, so we see that the clock with the relative
velocity will take longer to tick than if you were moving with the clock.
If this isn’t clear, we can also come to the same conclusion a different way. If we define
the speed of light as the amount of time it takes for the light beam to travel the distance
between the mirrors divided by the time it took to travel that distance, we see that:
c = 2L/At
but because the speed of light must be held constant in all frames of reference via locality,
we also have:
c = 2H/At
For the two preceding equations to be equivalent, At must be different for each system.
This means that if I were in a rocket moving at high velocity past you as you read this
book, you would look at the clock on my rocket wall and see it ticking more slowly than
your clock. Now, it may seem as though I would look out of my rocket and see your
clock running fast, but in fact the opposite is true. I would consider myself at rest and
you speeding past me such that I would say your clock is running slowly. This may seem
counterintuitive, but think of it in the same way as visual perspective. If you are at a
great distance from me, then you appear to be small. That doesn’t imply that I would
appear to be huge to you!
Now the amount of dilation for a given velocity v is given by the Lorentz transformation:
Relativistic Time | 31
www.it-ebooks.info
At’ = yAt
where:
Y =
is called the Lorentzfactor.
For velocities approaching the speed of light, the effect of time dilation is dramatic.
Imagine if you had a twin sister. She boards a spaceship and is accelerated to three-
fourths the speed of light relative to you on Earth. Upon her return, according to her
clock, 20 years have elapsed since she left. However, due to time dilation you will have
aged 30 years. While this seems paradoxical given that you both will have observed each
other’s time as running slowly compared to your own, the paradox is resolved by the
fact that special relativity claims only that inertial frames of reference are identical. For
the spaceship to return to Earth, it must accelerate by changing directions or, in other
words, become a noninertial frame of reference. Once the ship is no longer an inertial
frame of reference, there will appear to be a jump in the stay-at-home twin’s age. Strange
but true!
In addition to time dilation due to relative velocity, time also slows down in the presence
of strong gravitational fields as a result of Einstein’s theory of general relativity. This sort
of time dilation is not relative in the sense that if I were closer to a black hole than you,
we’d both agree that my clock is ticking less often than yours. However, given that all
physical processes would be slower, there is no way to establish which clock is “faster”
and which is “slower.” It is all relative!
As some games require a way to speed up or slow down time, the applications of time
dilation allow for a physical phenomenon to enable time manipulation. Imagine your
character needing to be sent to the future to complete some task. That character could
be put into a centrifuge and accelerated near the speed of light. Upon exiting the cen¬
trifuge, he would essentially have time-traveled into the future. However, if you plan to
stay within the limits of physics, then it is a one-way trip: there is no way to reverse time
in relativity! Additionally, if you would like to send a character to a nearby star, you can
do so within the limits of physics as well. By accelerating a spaceship, a human being
would be able to travel a great distance in a lifetime. In fact, with a constant acceleration
of 9.8 m/s 2 , which would make the astronauts feel like they were on Earth, you could
travel the entire visible galaxy. However, you would have essentially time-traveled bil¬
lions of years into the future. We are sure that you can imagine any number of scenarios
where such a mechanism might make your game more interesting while always re¬
maining in the realm of the physically possible!
32 | Chapter 1: Basic Concepts
www.it-ebooks.info
Now, besides those implications to games involving space flight or high-velocity travel,
time dilation is also important to some surprising digital electronic applications. For
instance, the Global Positioning System (GPS), described in detail in Chapter 22, must
take relativistic time dilation into account when calculating position. The satellite’s high
speed slows the clock compared to your watch on Earth; however, being farther up the
Earth’s gravity causes it to tick faster than a terrestrial clock. The specifics of this com¬
bined effect are discussed in Chapter 22.
Another point you might find interesting is that it is now easy to see how the “you can’t
travel faster than light” rule is a result of the theory of relativity. Should you accelerate
such that your velocity, v, is equal to c, the Lorentz transformation attempts to divide
by zero. For games where faster-than-light travel is a practical necessity, you will have
to imagine a mechanism to prevent this but be able to break the rules with style.
Relativistic Time | 33
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 2
Kinematics
In this chapter we’ll explain the fundamental aspects of the subject of kinematics.
Specifically, we’ll explain the concepts of linear and angular displacement, velocity, and
acceleration. We’ve prepared an example program for this chapter that shows you how
to implement the kinematic equations for particle motion. After discussing particle
motion, we go on to explain the specific aspects of rigid-body motion. This chapter,
along with the next chapter on force, is prerequisite to understanding the subject of
kinetics, which you’ll study in Chapter 4.
In the preface, we told you that kinematics is the study of the motion of bodies without
regard to the forces acting on the body. Therefore, in kinematics, attention is focused
on position, velocity, and acceleration of a body, how these properties are related, and
how they change over time.
Here you’ll look at two types of bodies, particles and rigid bodies. A rigid body is a
system of particles that remain at fixed distances from one another with no relative
translation or rotation among them. In other words, a rigid body does not change its
shape as it moves—or any changes in its shape are so small or unimportant that they
can safely be neglected. When you are considering a rigid body, its dimensions and
orientation are important, and you must account for both the body’s linear motion and
its angular motion.
A particle, on the other hand, is a body that has mass but whose dimensions are negli¬
gible or unimportant in the problem being investigated. For example, when considering
the path of a projectile or a rocket over a great distance, you can safely ignore the body’s
dimensions when analyzing its trajectory. When you are considering a particle, its linear
motion is important, but the angular motion of the particle itself is not. Think of it this
way: when looking at a particle, you are zooming way out to view the big picture, so to
speak, as opposed to zooming in as you do when looking at the rotation of rigid bodies.
35
www.it-ebooks.info
Whether you are looking at problems involving particles or rigid bodies, there are some
important kinematic properties common to both. These are, of course, the objects po¬
sition, velocity, and acceleration. The next section discusses these properties in detail.
Velocity and Acceleration
In general, velocity is a vector quantity that has magnitude and direction. The magnitude
of velocity is speed. Speed is a familiar term—it’s how fast your speedometer says you’re
going when driving your car down the highway. Formally, speed is the rate of travel, or
the ratio of distance traveled to the time it took to travel that distance. In math terms,
you can write:
v = As/At
where v is speed, the magnitude of velocity v, and As is distance traveled over the time
interval At. Note that this relation reveals that the units for speed are composed of the
basic dimension’s length divided by time, LIT. Some common units for speed are meters
per second, m/s; feet per second, ft/sec; and miles per hour, mi/hr.
Here’s a simple example (illustrated in Figure 2-1): a car is driving down a straight road
and passes marker one at time t l and marker two at time f 2 , where t 1 equals 0 seconds
and t 2 equals 1.136 seconds. The distance between these two markers, s, is 30 m. Calculate
the speed of the car.
1 1
V
CL
©
o
Figure 2-1. Example car speed
You are given that s equals 30 m; therefore, As equals 30 m and A, equals t 2 - f, or 1.136
seconds. The speed of the car over this distance is:
v = As/At = 30 m/1.136 sec = 26.4 m/sec
which is approximately 60 mi/hr. This is a simple one-dimensional example, but it brings
up an important point, which is that the speed just calculated is the average speed of the
car over that distance. You don’t know anything at this point about the car’s acceleration,
36 | Chapter 2: Kinematics
www.it-ebooks.info
or whether or not it is traveling at a constant 60 mi/hr. It could very well be that the car
was accelerating (or decelerating) over that 30 m distance.
To more precisely analyze the motion of the car in this example, you need to understand
the concept of instantaneous velocity. Instantaneous velocity is the specific velocity at
a given instant in time, not over a large time interval as in the car example. This means
that you need to look at very small At s. In math terms, you must consider the limit as
At approaches 0—that is, as At gets infinitesimally small. This is written as follows:
v = limAt-»o (As/At)
In differential terms, velocity is the derivative of displacement (change in position) with
respect to time:
v = ds/dt
You can rearrange this relationship and integrate over the intervals from s, to s 2 and f,
to t 2 , as shown here:
v dt = ds
/(si to s2) ds = /(ti to t2) V dt
s2 - si = As = / (tl t o t2) v dt
This relation shows that displacement is the integral of velocity over time. This gives
you a way of working back and forth between displacement and velocity.
Kinematics makes an important distinction between displacement and distance trav¬
eled. In one dimension, displacement is the same as distance traveled; however, with
vectors in space, displacement is actually the vector from the initial position to the final
position without regard to the path traveled, while displacement is the difference be¬
tween the starting position coordinates and the ending position coordinates. Thus, you
need to be careful when calculating average velocity given displacement if the path from
the starting position to the final position is not a straight line. When At is very small (as
it approaches 0), displacement and distance traveled are the same.
Another important kinematic property is acceleration, which should also be familiar to
you. Referring to your driving experience, you know that acceleration is the rate at which
you can increase your speed. Your friend who boasts that his brand new XYZ 2011 can
go from 0 to 60 in 4.2 seconds is referring to acceleration. Specifically, he is referring to
average acceleration.
Formally, average acceleration is the rate of change in velocity, or Av over At:
a = Av/At
Velocity and Acceleration | 37
www.it-ebooks.info
Taking the limit as A t goes to 0 gives the instantaneous acceleration:
a = limAt-^o Av/At
a = dv/dt
Thus, acceleration is the time rate of change in velocity, or, the derivative of velocity
with respect to time.
Multiplying both sides by dt and integrating yields:
dv = a dt
/(vl to v2) dv = / (t i to t2) a dt
v2 - vl = Av = /(u to t2) a dt
This relationship provides a means to work back and forth between velocity and accel¬
eration.
Thus, the relationships between displacement, velocity, and acceleration are:
a = dv/dt = d 2 s/dt 2
and:
v dv = a ds
This is the kinematic differential equation of motion (see the sidebar “Second Deriva¬
tives” on page 38 for some helpful background). In the next few sections you’ll see some
examples of the application of these equations for some common classes of problems
in kinematics.
Second Derivatives
In Chapter 1, we explained that you need not worry about the use of derivatives and
integrals in this book if you’re unfamiliar with calculus since we’ll show you how to
implement the code to take care of them computationally. That still applies, but here
we’ve introduced new notation:
d 2 s/dt 2
which is an equation representing acceleration as the second derivative of distance trav¬
eled with respect to time. You can think of second derivatives as just two successive
derivatives in the manner we explained in Chapter 1. In the case of distance traveled,
velocity, and acceleration, the first derivative of distance traveled with respect to time is
38 | Chapter 2: Kinematics
www.it-ebooks.info
velocity and the second derivative of distance traveled with respect to time is accelera¬
tion, which is the same as the first derivative of velocity with respect to time.
Constant Acceleration
One of the simplest classes of problems in kinematics involves constant acceleration. A
good example of this sort of problem involves the acceleration due to gravity, g, on
objects moving relatively near the earth’s surface, where the gravitational acceleration
is a constant 9.81 m/s 2 . Having constant acceleration makes integration over time rel¬
atively easy since you can pull the acceleration constant out of the integrand, leaving
just dt.
Integrating the relationship between velocity and acceleration described earlier when
acceleration is constant yields the following equation for instantaneous velocity:
/(vl to v2) dv = /(ti to t2) a dt
/(vl to v2) dv = a /(ti to t2) dt
v 2 — v i = a /(ti to t 2 ) dt
v 2 -vi = a (t 2 - t|)
v 2 = a t 2 — a ti + vi
When t l equals 0, you can rewrite this equation in the following form:
v 2 = a t 2 + V!
v 2 = vi + a t 2
This simple equation allows you to calculate the instantaneous velocity at any given time
by knowing the elapsed time, the initial velocity, and the constant acceleration.
You can also derive an equation for velocity as a function of displacement instead of
time by considering the kinematic differential equation of motion:
v dv = a ds
Integrating both sides of this equation yields the following alternative function for in¬
stantaneous velocity:
/(vl to v2) v dv = a / s i s2) ds
(v 2 2 - vi 2 ) / 2 = a (s 2 - si)
v 2 2 = 2a (s 2 - si) + vi 2
Constant Acceleration | 39
www.it-ebooks.info
You can derive a similar formula for displacement as a function of velocity, acceleration,
and time by integrating the differential equation:
v dt = ds
with the formula derived earlier for instantaneous velocity:
v 2 — v i + at
substituted for v. Doing so yields the formula:
s 2 = s l + v i t + (a t 2 ) /2
In summary, the three preceding kinematic equations derived are:
v 2 = vi + a t 2
v 2 2 = 2a (s 2 - si) + vi 2
s 2 = si + vi t + (a t 2 ) / 2
Remember, these equations are valid only when acceleration is constant. Note that ac¬
celeration can be 0 or even negative in cases where the body is decelerating.
You can rearrange these equations by algebraically solving for different variables, and
you can also derive other handy equations using the approach that we just demonstrated.
For your convenience, we’ve provided some other useful kinematic equations for con¬
stant acceleration problems in Table 2-1.
Table 2-1. Constant acceleration kinematic formulas
To find:
Given these:
Use this:
a
At, v 1( v 2
a = (v 2 —v n ) / At
a
At, v 1( As
a = (2 As — 2 v 2 At) / (At) 2
a
v-i, v 2 , As
a = (v 2 2 — v 2 2 ) / (2 As)
As
a,v„v 2
As = (v 2 2 - v 2 2 ) / (2a)
As
At, v,, v 2
As = (At/2) (v, + v 2 )
At
a,v„v 2
At = (v 2 -v,)/a
At
a, v„ As
At =
[fvf + laAs - vj la
At
v„ v 2 , As
At=(2As)/(v 1 + v 2 )
Vl
At, a, v 2
v n = v 2 — aAt
Vl
At, a, As
v, = As/At - (aAt)/2
40 | Chapter 2: Kinematics
www.it-ebooks.info
To find: Given these: Use this:
Vi a, v 2 , As v,=_
Jvf-2aAs
In cases where acceleration is not constant, but is some function of time, velocity, or
position, you can substitute the function for acceleration into the differential equations
shown earlier to derive new equations for instantaneous velocity and displacement. The
next section considers such a problem.
Nonconstant Acceleration
A common situation that arises in real-world problems is when drag forces act on a
body in motion. Typically, drag forces are proportional to velocity squared. Recalling
the equation of Newton’s second law of motion, F = ma, you can deduce that the accel¬
eration induced by these drag forces is also proportional to velocity squared
Later we’ll show you some techniques to calculate this sort of drag force, but for now
let the functional form of drag-induced acceleration be:
a = -kv 2
where k is a constant and the negative sign indicates that this acceleration acts in the
direction opposing the body’s velocity. Now substituting this formula for acceleration
into the previous equation and then rearranging yields:
a = dv/dt
-kv 2 = dv/dt
-k dt = dv/v 2
If you integrate the right side of this equation from v x to v 2 and the left side from 0 to
t, and then solve for v 2 , you’ll get this formula for the instantaneous velocity as a function
of the initial velocity and time:
-k /(o to t) dt = /(vi to v2) (1/v 2 ) dv
-k t = 1/vj - 1 /V 2
v 2 = vi /(I + vikt)
If you substitute this equation for v in the relation v = ds/dt and integrate again, you’ll
end up with a new equation for displacement as a function of initial velocity and time;
see the following procedure:
v dt = ds; where v = vj / (1 + vjk t)
Nonconstant Acceleration | 41
www.it-ebooks.info
If s t equals 0, then:
/( 0 to t) v dt = /(si to s2) ds
/(0 to t) [vi / (1 + vik t)] dt = /( S i t0 s 2 ) ds
ln(l + vi k t) / k = S 2 - si
s = ln(l + vi k t) /k
Note that in this equation, In is the natural logarithm operator.
This example demonstrates the relative complexity of nonconstant acceleration prob¬
lems versus constant acceleration problems. It’s a fairly simple example where you are
able to derive closed-form equations for velocity and displacement. In practice, however,
there may be several different types of forces acting on a given body in motion, which
could make the expression for induced acceleration quite complicated. This complexity
would render a closed-form solution like the preceding one impossible to obtain unless
you impose some simplifying restrictions on the problem, forcing you to rely on other
solution techniques like numerical integration. We’ll talk about this sort of problem in
greater depth in Chapter 11.
2D Particle Kinematics
When we are considering motion in one dimension—that is, when the motion is re¬
stricted to a straight line—it is easy enough to directly apply the formulas derived earlier
to determine instantaneous velocity, acceleration, and displacement. However, in two
dimensions, with motion possible in any direction on a given plane, you must consider
the kinematic properties of velocity, acceleration, and displacement as vectors.
Using rectangular coordinates in the standard Cartesian coordinate system, you must
account for the x and y components of displacement, velocity, and acceleration. Essen¬
tially, you can treat the x and y components separately and then superimpose these
components to define the corresponding vector quantities.
To help keep track of these x and y components, let i and j be unit vectors in the x- and
y-directions, respectively. Now you can write the kinematic property vectors in terms
of their components as follows:
V = v x i +Vyj
a = a x i + a y j
If x is the displacement in the x-direction and y is the displacement in the y-direction,
then the displacement vector is:
s = x i + y j
42 | Chapter 2: Kinematics
www.it-ebooks.info
It follows that:
v = ds/dt = dx/dt i + dy/dt j
a = dv/dt = d 2 s/dt 2 = d 2 x/dt 2 i + d 2 y/dt 2 j
Consider a simple example where you’re writing a shooting game and you need to figure
out the vertical drop in a fired bullet from its aim point to the point at which it actually
hits the target. In this example, assume that there is no wind and no drag on the bullet
as it flies through the air (we’ll deal with wind and drag on projectiles in Chapter 6).
These assumptions reduce the problem to one of constant acceleration, which in this
case is that due to gravity. It is this gravitational acceleration that is responsible for the
drop in the bullet as it travels from the rifle to the target. Figure 2-2 illustrates the
problem.
^ I Aim point
—.c
f rj
Hit point
-- n -H
1
Figure 2-2. A 2D kinematics example problem
tfi*
-
While we’re talking about guns and shooting here, we should point out
*<?; j that these techniques can be applied just as easily to simulating the flight
— . of angry birds in Angry Birds being shot from oversized slingshots, as
in the very popular iPhone app. Heck, you can use these techniques to
simulate flying monkeys, ballistic shoes, or coconuts being hurled at
Navy combatants! This particle kinematic stuff is perfect for diversion¬
ary smartphone apps.
Let the origin of the 2D coordinate system be at the end of the rifle with the x-axis
pointing toward the target and the y-axis pointing up. Positive displacements along the
x-axis are toward the target, and positive displacements along the y-axis are upward.
This implies that the gravitational acceleration will act in the negative y-direction.
Treating the x and y components separately allows you to break up the problem into
small, easy-to-manage pieces. Looking at the x component first, you know that the bullet
will leave the rifle with an initial muzzle velocity v m in the x-direction, and since we are
neglecting drag, this speed will be constant. Thus:
2D Particle Kinematics | 43
www.it-ebooks.info
a x = 0
v x = v m
X = v x t = v m t
Now looking at they component, you know that the initial speed in the y-direction, as
the bullet leaves the rifle, is 0, but the y-acceleration is -g (due to gravity). Thus:
a y = -g = dv y /dt
v y = a y t = —g t
y = (1/2) a y t 2 = -(1/2) g t 2
The displacement, velocity, and acceleration vectors can now be written as:
s = (v m t) i — (1/2 g t 2 ) j
v = (v m ) i - (g t) j
a = - (g) j
These equations give the instantaneous displacement, velocity, and acceleration for any
given instant between the time the bullet leaves the rifle and the time it hits the target.
The magnitudes of these vectors give the total displacement, velocity, and acceleration
at a given time. For example:
S = J(v m t)~ + (l/2gt 2 ) 2
V = V(v m ) 2 + (gt) 2
a = 4g~ 2 = g
To calculate the bullets vertical drop at the instant the bullet hits the target, you must
first calculate the time required to reach the target; then, you can use that time to cal¬
culate they component of displacement, which is the vertical drop. Here are the formulas
to use:
thit = XhiAm = n/v m
d = Yhit = -(1/2) g (thit) 2
where n is the distance from the rifle to the target and d is the vertical drop of the bullet
at the target.
44 | Chapter 2: Kinematics
www.it-ebooks.info
If the distance to the target, n, equals 500 m and the muzzle velocity, v m , equals 800 m/
sec, then the equations for ^hit and d give:
thit = 0.625 sec
d = 1.9 m
These results tell you that in order to hit the intended target at that range, you’ll need
to aim for a point about 2 m above it.
3D Particle Kinematics
Extending the kinematic property vectors to three dimensions is not very difficult. It
simply involves the addition of one more component to the vector representations
shown in the previous section on 2D kinematics. Introducing k as the unit vector in the
z-direction, you can now write:
s = xi + yj + zk
v = ds/dt = dx/dt i + dy/dt j + dz/dt k
a = d 2 s/dt 2 = d 2 x/dt 2 i + d 2 y/dt 2 j + d 2 z/dt 2 k
Instead of treating two components separately and then superimposing them, you now
treat three components separately and superimpose these. This is best illustrated by an
example.
Suppose that instead of a hunting game, you’re now writing a game that involves the
firing of a cannon from, say, a battleship to a target some distance away—for example,
another ship or an inland target like a building. To add complexity to this activity for
your user, you’ll want to give her control of several factors that affect the shell’s trajectory
—namely, the firing angle of the cannon, both horizontal and vertical angles, and the
muzzle velocity of the shell, which is controlled by the amount of powder packed behind
the shell when it’s loaded into the cannon. The situation is set up in Figure 2-3.
Figure 2-3. A 3D kinematics example problem
3D Particle Kinematics | 45
www.it-ebooks.info
We’ll show you how to set up the kinematic equations for this problem by treating each
vector component separately at first and then combining these components.
X Components
The x components here are similar to those in the previous section’s rifle example in
that there is no drag force acting on the shell; thus, the x component of acceleration is
0, which means that the x component of velocity is constant and equal to the x compo¬
nent of the muzzle velocity as the shell leaves the cannon. Note that since the cannon
barrel may not be horizontal, you’ll have to compute the x component of the muzzle
velocity, which is a function of the direction in which the cannon is aimed.
The muzzle velocity vector is:
v m = v mx i + v my j + v mz k
and you are given only the direction of v m as determined by the direction in which the
user points the cannon, and its magnitude as determined by the amount of powder the
user packs into the cannon. To calculate the components of the muzzle velocity, you
need to develop some equations for these components in terms of the direction angles
of the cannon and the magnitude of the muzzle velocity.
You can use the direction cosines of a vector to determine the velocity components as
follows:
cos 0 X = Vmx/Vm
COS 0y = V m y/v m
cox 9 Z = v mz /v m
Refer to Appendix A for a description and illustration of vector direction cosines.
Since the initial muzzle velocity vector direction is the same as the direction in which
the cannon is aimed, you can treat the cannon as a vector with a magnitude of L, its
length, and pointing in a direction defined by the angles given in this problem. Using
the cannon length, L, and its components instead of muzzle velocity in the equations
for direction cosines gives:
cos 0 X = L x /L
COS 9y = Ly/L
cos 0 Z = L z /L
In this example, you are given the angles a and y (see Figure 2-4) that define the cannon
orientation.
46 | Chapter 2: Kinematics
www.it-ebooks.info
Figure 2-4. Cannon orientation
Using these angles, it follows that the projection, b, of the cannon length, L, onto the x-
z plane is:
b = L cos(90° - a)
and the components of the cannon length, L, on each coordinate axis are:
Lx = b cos y
Ly = L cos a
Lz = b sin y
Now that you have the information required to compute direction cosines, you can write
equations for the initial muzzle velocity components as follows:
v mx = v m cos 0 X
v my = v m cos 9y
v mz = v m cos
Finally, you can write the x components of displacement, velocity, and acceleration as
follows:
a x = 0
v x = v mx = v m cos 0 X
x = v x t = (v m cos 0 X ) t
Y Components
They components are just like the previous rifle example, again with the exception here
of the initial velocity in the y-direction:
3D Particle Kinematics | 47
www.it-ebooks.info
Vmy = v m COS 0 y
Thus:
a y — -g
v y = v my + a t = (v m cos 0 y ) - g t
Before writing the equation for they component of displacement, you need to consider
the elevation of the base of the cannon, plus the height of the end of the cannon barrel,
in order to calculate the initial y component of displacement when the shell leaves the
cannon. Let y h be the elevation of the base of the cannon, and let L be the length of the
cannon barrel; then the initial y component of displacement, y Q , is:
y 0 = Yb + L cos a
Now you can write the equation for y as:
y = y 0 + v my t + (1/2) at 2
y = (yb + L cos a) + (v m cos 0 y ) t - (1/2) g t 2
Z Components
The z components are largely analogous to the x components and can be written as
follows:
a z = 0
v z = V m z = v m COS 0 Z
Z = V z t = (v m COS 0 Z ) t
The Vectors
With the components all worked out, you can now combine them to form the vector
for each kinematic property. Doing so for this example gives the displacement, velocity,
and acceleration vectors shown here:
s = [(v m cos 0 X ) t] i + [(yb + L cos a) + (v m cos 0 y ) t - (1/2) g t 2 ]
j + [(v m cos 0 Z ) t ] k
v = [v m cos 0 X ] i + [(v m cos 0 y ) - g t ] j + [v m cos 0 Z ] k
a — -g j
48 | Chapter 2: Kinematics
www.it-ebooks.info
Observe here that the displacement vector essentially gives the position of the shells
center of mass at any given instant in time; thus, you can use this vector to plot the shells
trajectory from the cannon to the target.
Hitting the Target
Now that you have the equations fully describing the shell’s trajectory, you need to
consider the location of the target in order to determine when a direct hit occurs. To
show you how to do this, we’ve prepared a sample program that implements these
kinematic equations along with a simple bounding box collision detection method for
checking whether or not the shell has struck the target. Basically, at each time step where
we calculate the position of the shell after it has left the cannon, we check to see if this
position falls within the bounding dimensions of the target obj ect represented by a cube.
The sample program is set up such that you can change all of the variables in the sim¬
ulation and view the effects of your changes. Figure 2-5 shows the main screen for the
cannon example program, with the governing variables shown on the left. The upper
illustration is a bird’s-eye view looking down on the cannon and the target, while the
lower illustration is a profile (side) view.
Figure 2-5. Cannon sample program main window
3D Particle Kinematics | 49
www.it-ebooks.info
You can change any of the variables shown on the main window and press the Fire
button to see the resulting flight path of the shell. A message box will appear when you
hit the target or when the shell hits the ground. The program is set up so you can
repeatedly change the variables and press Fire to see the result without erasing the
previous trial. This allows you to gauge how much you need to adjust each variable in
order to hit the target. Press the Refresh button to redraw the views when they get too
cluttered.
Figure 2-6 shows a few trial shots that we made before finally hitting the target.
in
* *
Figure 2-6. Trial shots (profile view)
The code for this example is really quite simple. Aside from the overhead of the window,
controls, and illustrations setup, all of the action takes place when the Fire button is
pressed. In pseudocode, the Fire button’s pressed event handler looks something like
this:
FIRE BUTTON PRESSED EVENT:
Fetch and store user input values for global variables,
Vn, Alpha, Ganna, L, Yb, X, Y, Z, Length, Width, Height...
Initialize the tine and status variables...
status = 0;
tine = 0;
Start stepping through tine for the sinulation
until the target is hit, the shell hits
the ground, or the sinulation tines out...
while(status == 0)
{
//do the next tine step
status = DoSinulationQ;
50 | Chapter 2: Kinematics
www.it-ebooks.info
Update the display...
}
// Report results
if (status == 1)
Display DIRECT HIT message to the user...
if (status == 2)
Display MISSED TARGET message to the user...
if (status == 3)
Display SIMULATION TIMED OUT message to the user...
The first task is to simply get the new values for the variables shown on the main window.
After that, the program enters a while loop, stepping through increments of time and
recalculating the position of the shell projectile using the formula for the displacement
vector, s, shown earlier. The shell position at the current time is calculated in the function
DoSirnulation. Immediately after calling DoSimulation, the program updates the il¬
lustrations on the main window, showing the shell’s trajectory. DoSimulation returns
0, keeping the while loop going, if there has not yet been a collision or if the time has
not yet reached the preset time-out value.
Once the while loop terminates by DoSimulation returning nonzero, the program
checks the return value from this function call to see if a hit has occurred between the
shell and the ground or the shell and the target. Just so the program does not get stuck
in this while loop, DoSimulation will return a value of 3, indicating that it is taking too
long.
Now let’s look at what’s happing in the function DoSimulation (we’ve also included here
the global variables that are used in DoSimulation).
//-.//
// Define a custom type to represent
// the three components of a 3D vector, where
// i represents the x component, j represents
// the y component, and k represents the z
// component
//- ... //
typedef struct TVectorTag
{
double i;
double j;
double k;
} TVector;
//.//
// Now define the variables required for this simulation
//.//
double Vm; // Magnitude of muzzle velocity, m/s
3D Particle Kinematics | 51
www.it-ebooks.info
double
Alpha;
//
//
//
//
double
Gamma;
//
//
//
//
double
L;
//
double
Yb;
//
double
X;
//
double
Y;
//
double
Z;
//
double
Length:
; //
double
Width;
//
double
Height:
; //
TVector
s;
//
double
time;
//
//
double
tine;
//
//
double
g;
//
Angle from y-axls (upward) to the cannon.
When this angle is 0, the cannon is pointing
straight up, when it is 90 degrees, the cannon
is horizontal
Angle from x-axis, in the x-z plane to the cannon.
When this angle is 0, the cannon is pointing in
the positive x-direction, positive values of this angle
are toward the positive z-axis
This is the length of the cannon, m
This is the base elevation of the cannon, m
The x-position of the center of the target, m
The y-position of the center of the target, m
The z-position of the center of the target, m
The length of the target measured along the x-axis, m
The width of the target measured along the z-axis, m
The height of the target measure along the y-axis, m
The shell position (displacement) vector
The time from the instant the shell leaves
the cannon, seconds
The time increment to use when stepping through
the simulation, seconds
acceleration due to gravity, m/s A 2
//.-//
// This function steps the simulation ahead in time. This is where the kinematic
// properties are calculated. The function will return 1 when the target is hit,
// and 2 when the shell hits the ground (x-z plane) before hitting the target;
// otherwise, the function returns 0.
//.-//
int DoSimulation(void)
//.-//
{
double cosX;
double cosY;
double cosZ;
double xe, ze;
double b. Lx, Ly, Lz;
double txl, tx2, tyl, ty2, tzl, tz2;
// step to the next time in the simulation
time+=tlnc;
// First calculate the direction cosines for the cannon orientation.
// In a real game, you would not want to put this calculation in this
// function since it is a waste of CPU time to calculate these values
// at each time step as they never change during the sim. We only put them
// here in this case so you can see all the calculation steps in a single
// function.
52 | Chapter 2: Kinematics
www.it-ebooks.info
b = L * cos((90-Alpha) *3.14/180); // projection of barrel onto x-z plane
Lx = b * cos(Gamma * 3.14/180); // x-component of barrel length
Ly = L * cos(Alpha * 3.14/180); // y-component of barrel length
Lz = b * sin(Gamma * 3.14/180); // z-conponent of barrel length
cosX = Lx/L;
cosY = Ly/L;
cosZ = Lz/L;
// These are the x and z coordinates of the very end of the cannon barrel
// we'll use these as the initial x and z displacements
xe = L * cos((90-Alpha) *3.14/180) * cos(Gamma * 3.14/180);
ze = L * cos((90-Alpha) *3.14/180) * sin(Gamma * 3.14/180);
// Now we can calculate the position vector at this time
s.i = Vm * cosX * time + xe;
s.j = (Yb + L * cos(Alpha*3.14/180)) + (Vm * cosY * time) -
(0.5 * g * time * time);
s.k = Vm * cosZ * time + ze;
// Check for collision with target
// Get extents (bounding coordinates) of the target
txl = X - Length/2;
tx2 = X + Length/2;
tyl = Y - Height/2;
ty2 = Y + Height/2;
tzl = Z - Width/2;
tz2 = Z + Width/2;
// Now check to see if the shell has passed through the target
// We're using a rudimentary collision detection scheme here where
// we simply check to see if the shell's coordinates are within the
// bounding box of the target. This works for demo purposes, but
// a practical problem is that you may miss a collision if for a given
// time step the shell's change in position is large enough to allow
// it to "skip" over the target.
// A better approach is to look at the previous time step's position data
// and to check the line from the previous position to the current position
// to see if that line intersects the target bounding box.
if( (s.i >= txl && s.i <= tx2) &&
(s.j >= tyl && s.j <= ty2) &&
(s.k >= tzl && s.k <= tz2) )
return 1;
// Check for collision with ground (x-z plane)
if(s.j <= 0)
return 2;
// Cut off the simulation if it's taking too long
// This is so the program does not get stuck in the while loop
if(time>3600)
return 3;
3D Particle Kinematics | 53
www.it-ebooks.info
return 0;
}
We’ve commented the code so that you can readily see what’s going on. This function
essentially does four things: 1) increments the time variable by the specified time in¬
crement, 2) calculates the initial muzzle velocity components in the x-, y-, and z-
directions, 3) calculates the shell’s new position, and 4) checks for a collision with the
target using a bounding box scheme or the ground.
Here’s the code that computes the shell’s position:
// Now we can calculate the position vector at this tine
s.i = Vm * cosX * tine + xe;
s.j = (Yb + L * cos(Alpha*3.14/180)) + (Vn * cosY * tine) -
(0.5 * g * tine * tine);
s.k = Vn * cosZ * tine + ze;
This code calculates the three components of the displacement vector, s, using the for¬
mulas that we gave you earlier. If you wanted to compute the velocity and acceleration
vectors as well, just to see their values, you should do so in this section of the program.
You can set up a couple of new global variables to represent the velocity and acceleration
vectors, just as we did with the displacement vector, and apply the velocity and accel¬
eration formulas that we gave you.
That’s all there is to it. It’s obvious by playing with this sample program that the shell’s
trajectory is parabolic in shape, which is typical projectile motion. We’ll take a more
detailed look at this sort of motion in Chapter 6.
Even though we put a comment in the source code, we must reiterate a warning here
regarding the collision detection scheme that we used in this example. Because we’re
checking only the current position coordinate to see if it falls within the bounding
dimensions of the target cube, we run the risk of skipping over the target if the change
in position is too large for a given time step. A better approach would be to keep track
of the shell’s previous position and check to see if the line connecting the previous
position to the new one intersects the target cube.
Kinematic Particle Explosion
At this point you might be wondering how particle kinematics can help you create
realistic game content unless you’re writing a game that involves shooting a gun or a
cannon. If so, let us offer you a few ideas and then show you an example. Say you’re
writing a football simulation game. You can use particle kinematics to model the tra¬
jectory of the football after it’s thrown or kicked. You can also treat the wide receivers
as particles when calculating whether or not they’ll be able to catch the thrown ball. In
this scenario you’ll have two particles—the receiver and the ball—traveling independ¬
ently, and you’ll have to calculate when a collision occurs between these two particles,
54 | Chapter 2: Kinematics
www.it-ebooks.info
indicating a catch (unless, of course, your player is all thumbs and fumbles the ball after
it hits his hands). You can find similar applications for other sports-based games as well.
What about a 3D “shoot ’em up” game? How could you use particle kinematics in this
genre aside from bullets, cannons, grenades, and the like? Well, you could use particle
kinematics to model your player when she jumps into the air, either from a run or from
a standing position. For example, your player reaches the middle of a catwalk only to
find a section missing, so you immediately back up a few paces to get a running head
start before leaping into the air, hoping to clear the gap. This long-jump scenario is
perfect for using particle kinematics. All you really need to do is define your players
initial velocity, both speed and take-off angle, and then apply the vector formula for
displacement to calculate whether or not the player makes the jump. You can also use
the displacement formula to calculate the player’s trajectory so that you can move the
player’s viewpoint accordingly, giving the illusion of leaping into the air. You may in fact
already be using these principles to model this action in your games, or at least you’ve
seen it done if you play games of this genre. If your player happens to fall short on the
j ump, you can use the formulas for velocity to calculate the player’s impact velocity when
she hits the ground below. Based on this impact velocity, you can determine an appro¬
priate amount of damage to deduct from the player’s health score, or if the velocity is
over a certain threshold, you can say goodbye to your would-be adventurer!
Another use for simple particle kinematics is for certain special effects like particle
explosions. This sort of effect is quite simple to implement and really adds a sense of
realism to explosion effects. The particles don’t just fly off in random, straight-line tra¬
jectories. Instead, they rise and fall under the influence of their initial velocity, angle,
and the acceleration due to gravity, which gives the impression that the particles have
mass.
So, let’s explore an example of a kinematic particle explosion. The code for this example
is taken from the cannon example discussed previously, so a lot of it should look familiar
to you. Figure 2-7 shows this example program’s main window.
Kinematic Particle Explosion | 55
www.it-ebooks.info
Figure 2-7. Particle explosion example program
The explosion effect takes place in the large rectangular area on the right. While the
black dots representing exploding particles are certainly static in the figure, we assure
you they move in the most spectacular way during the simulation.
In the edit controls on the left, you specify an x- and y-position for the effect, along with
the initial velocity of the particles (which is a measure of the explosion’s strength), a
duration in milliseconds, a gravity factor, and finally an angle. The angle parameter can
be any number between 0 and 360 degrees or 999. When you specify an angle in the
range of 0 to 360 degrees, all the particles in the explosion will be launched generally in
that direction. If you specify a value of999, then all the particles will shoot off in random
directions. The duration parameter is essentially the life of the effect. The particles will
fade out as they approach that life.
The first thing you need to do for this example is set up some structures and global
variables to represent the particle effect and the individual particles making up the effect
along with the initial parameters describing the behavior of the effect as discussed in
the previous paragraph. Here’s the code:
//.//
// Define a custom type to represent each particle in the effect.
//.//
typedef struct _TParticle
{
float x; // x coordinate of the particle
56 | Chapter 2: Kinematics
www.it-ebooks.info
float
y;
// y coordinate of the particle
float
vi;
// initial velocity
float
angle;
// initial trajectory (direction)
int
life;
// duration in milliseconds
int
r;
// red component of particle's color
int
g;
// green component of particle's color
int
b;
// blue component of particle's color
int
time;
// keeps track of the effect's time
float
gravity;
// gravity factor
BOOL
Active;
// indicates whether this particle
// is active or dead
} TPartlcle;
#define _MAXPARTICLES 50
typedef struct -TParticleExplosion
{
TPartlcle p[_MAXPARTICLES]; // list of particles
// making up this effect
int V0; // initial velocity, or strength, of the effect
int x; // initial x location
int y; // initial y location
BOOL Active; // indicates whether this effect is
//active or dead
} TParticleExplosion;
//.//
// Now define the variables required for this simulation
//...//
TParticleExplosion Explosion;
int xc; // x coordinate of the effect
int yc; // y coordinate of the effect
int V0; // initial velocity
int Duration; // life in milliseconds
float Gravity; // gravity factor (acceleration)
float Angle; // indicates particles' direction
You can see from this code that the particle explosion effect is made up of a collection
of particles. The behavior of each particle is determined by kinematics and the initial
parameters set for each particle.
Whenever you press the Go button, the initial parameters that you specified are used
to initialize the particle explosion (if you press the Random button, the program ran¬
domly selects these initial values for you). This takes place in the function called Crea
teParticleExplosion:
/////////////////////////////////////////////////////////////////////
/* This function creates a new particle explosion effect.
x,y: starting point of the effect
Vinit: a measure of how fast the particles will be sent flying
Kinematic Particle Explosion | 57
www.it-ebooks.info
(it's actually the initial velocity of the particles)
life: life of the particles in milliseconds; particles will
fade and die out as they approach
their specified life
gravity: the acceleration due to gravity, which controls the
rate at which particles will fall
as they fly
angle: initial trajectory angle of the particles,
specify 999 to create a particle explosion
that emits particles in all directions; otherwise,
0 right, 90 up, 180 left, etc...
*/
void CreateParticleExplosion(int x, int y, int Vinit, int life,
float gravity, float angle)
{
int i;
int m;
float f;
Explosion.Active = TRUE;
Explosion.x = x;
Explosion.y = y;
Explosion.V0 = Vinit;
for(i=0; i<_MAXPARTICLES; i++)
{
Explosion.p[i].x = 0;
Explosion.p[i].y = 0;
Explosion.p[i].vi = tb_Rnd(Vinit/2, Vinit);
tf(angle < 999)
{
if(tb_Rnd(0,l) == 0)
m = -1;
else
m = 1;
Explosion.p[i].angle = -angle + m * tb_Rnd(0,10);
} else
Explosion.p[i].angle = tb_Rnd(0,360);
f = (float) tb_Rnd(80, 100) / 100.0f;
Explosion.p[i].life = tb_Round(life * f);
Explosion.p[i].r = 255;//tb_Rnd(225, 255);
Explosion.p[i].g = 255;//tb_Rnd(85, 115);
Explosion.p[i].b = 255;//tb_Rnd(15, 45);
Explosion.p[i].time = 0;
Explosion.p[i].Active = TRUE;
Explosion.p[i].gravity = gravity;
}
}
58 | Chapter 2: Kinematics
www.it-ebooks.info
Here you can see that all the particles are set to start off in the same position, as specified
by the x andy coordinates that you provide; however, you’ll notice that the initial velocity
of each particle is actually randomly selected from a range of Vinit/2 to Vinit. We do
this to give the particle behavior some variety. We do the same thing for the life param¬
eter of each particle so they don’t all fade out and die at the exact same time.
After the particle explosion is created, the program enters a loop to propagate and draw
the effect. The loop is a while loop, as shown here in pseudocode:
whlle(status)
{
Clear the off screen buffer...
status = DrawParticleExplosion( );
Copy the off screen buffer to the screen...
}
The while loop continues as long as status remains true, which indicates that the
particle effect is still alive. After all the particles in the effect reach their set life, then the
effect is dead and status will be set to false. All the calculations for the particle behavior
actually take place in the function called DrawParticleExploslon; the rest of the code
in this while loop is for clearing the off-screen buffer and then copying it to the main
window.
DrawParticleExploslon updates the state of each particle in the effect by calling an¬
other function, UpdateParticleState, and then draws the effect to the off-screen buffer
passed in as a parameter. Here’s what these two functions look like:
//-.//
// Draws the particle system and updates the state of each particle.
// Returns false when all of the particles have died out.
//-.//
BOOL DrawParticleExplosion(void)
{
int i;
BOOL finished = TRUE;
float r;
if(Explosion.Active)
for(i=0; i<_MAXPARTICLES; i++)
{
if(Explosion.p[i].Active)
{
finished = FALSE;
// Calculate a color scale factor to fade the particle's color
// as its life expires
r = ((float)(Explosion.p[i].Life-
Explosion .p[i].time)/(float)(Explosion.p[i].life));
Kinematic Particle Explosion | 59
www.it-ebooks.info
Draw the particle as a small circle...
}
}
Explosion.p[l].Active = UpdatePartlcleState(&(Exploslon.p[l]),
10 );
If(finished)
Explosion.Active = FALSE;
return !finished;
}
//.//
/* This Is generic function to update the state of a given particle,
p: pointer to a particle structure
dtlme: time Increment In milliseconds to
advance the state of the particle
If the total elapsed time for this particle has exceeded the particle's
set life, then this function returns FALSE, Indicating that the particle
should expire.
*/
BOOL UpdatePartlcleState(TPartlcle* p, Int dtlme)
{
BOOL retval;
float t;
p->tlme+=dtlme;
t = (float)p->tlme/1000.0f;
p->x = p->vl * cos(p->angle*PI/180.0f) * t;
p->y = p->vl * sln(p->angle*PI/180.0f) * t + (p->gravlty*t*t)/2.0f;
If (p->tlme >= p->llfe)
retval = FALSE;
else
retval = TRUE;
return retval;
}
UpdateParticleState uses the kinematic formulas that we’ve already shown you to
update the particle’s position as a function of its initial velocity, time, and the acceleration
due to gravity. After UpdateParticleState is called, DrawParticleExplosion scales
down each particle’s color, fading it to black, based on the life of each particle and elapsed
time. The fade effect is simply to show the particles dying slowly over time instead of
disappearing from the screen. The effect resembles the behavior of fireworks as they
explode in the night sky.
60 | Chapter 2: Kinematics
www.it-ebooks.info
Rigid-Body Kinematics
The formulas for displacement, velocity, and acceleration discussed in the previous
sections apply equally well for rigid bodies as they do for particles. The difference is that
with rigid bodies, the point on the rigid body that you track, in terms of linear motion,
is the body’s center of mass (gravity).
When a rigid body translates with no rotation, all of the particles making up the rigid
body move on parallel paths since the body does not change its shape. Further, when a
rigid body does rotate, it generally rotates about axes that pass through its center of
mass, unless the body is hinged at some other point about which it’s forced to rotate.
These facts make the center of mass a convenient point to use to track its linear motion.
This is good news for you since you can use all of the material you learned on particle
kinematics here in your study of rigid-body kinematics.
The procedure for dealing with rigid bodies involves two distinct aspects: 1) tracking
the translation of the body’s center of mass, and 2) tracking the body’s rotation. The first
aspect is old hat by now—just treat the body as a particle. The second aspect, however,
requires you to consider a few more concepts—namely, local coordinates, angular dis¬
placement, angular velocity, and angular acceleration.
For most of the remainder of this chapter, we’ll discuss plane kinematics of rigid bodies.
Plane motion simply means that the body’s motion is restricted to a flat plane in space
where the only axis of rotation about which the body can rotate is perpendicular to the
plane. Plane motion is essentially two-dimensional. This allows us to focus on the new
kinematic concepts of angular displacement, velocity, and acceleration without getting
lost in the math required to describe arbitrary rotation in three dimensions.
You might be surprised by how many problems lend themselves to plane kinematic
solutions. For example, in some popular 3D “shoot ’em up” games, your character is able
to push objects, such as boxes and barrels, around on the floor. While the game world
here is three dimensions, these particular objects may be restricted to sliding on the
floor—a plane—and thus can be treated like a 2D problem. Even if the player pushes
on these objects at some angle instead of straight on, you’ll be able to simulate the sliding
and rotation of these objects using 2D kinematics (and kinetics) techniques.
Rigid-Body Kinematics | 61
www.it-ebooks.info
Local Coordinate Axes
Earlier, we defined the Cartesian coordinate system to use for your fixed global refer¬
ence, or world coordinates. This world coordinate system is all that’s required when
treating particles; however, for rigid bodies you’ll also use a set of local coordinates fixed
to the body. Specifically, this local coordinate system will be fixed at the body’s center-
of-mass location. You’ll use this coordinate system to track the orientation of the body
as it rotates.
For plane motion, we require only one scalar quantity to describe the body’s orientation.
This is illustrated in Figure 2-8.
Figure 2-8. Local coordinate axes
Here the orientation, Cl, is defined as the angular difference between the two sets of
coordinate axes: the fixed world axes and the local body axes. This is the so-called Euler
angle. In general 3D motion there is a total of three Euler angles, which are usually called
yaw, pitch, and roll in aerodynamic and hydrodynamic jargon. While these angular
representations are easy to visualize in terms of their physical meaning, they aren’t so
nice from a numerical point of view, so you’ll have to look for alternative representations
when writing your 3D real-time simulator. These issues are addressed in Chapter 9.
Angular Velocity and Acceleration
In two-dimensional plane motion, as the body rotates, Cl will change, and the rate at
which it changes is the angular velocity, w. Likewise, the rate at which to changes is the
angular acceleration, a. These angular properties are analogous to the linear properties
of displacement, velocity, and acceleration. The units for angular displacement, velocity,
62 | Chapter 2: Kinematics
www.it-ebooks.info
and acceleration are radians (rad), radians per sec (rad/s), and radians per second-
squared (rad/s 2 ), respectively.
Mathematically, you can write these relations between angular displacement, angular
velocity, and angular acceleration as:
co = d£2/dt
a = dco/dt = d 2 £2/dt 2
co = fa dt
Cl = ft o dt
co dco = a d£2
In fact, you can substitute the angular properties Cl, co, and a for the linear properties
s, v, and a in the equations derived earlier for particle kinematics to obtain similar
kinematic equations for rotation. For constant angular acceleration, you’ll end up with
the following equations:
(02 = ®i + a t
tt> 2 2 = uq 2 + 2 a (£^2 - Qi)
Cl 2 = + cojt + (1/2) a t“
When a rigid body rotates about a given axis, every point on the rigid body sweeps out
a circular path around the axis of rotation. You can think of the body’s rotation as causing
additional linear motion of each particle making up the body—that is, this linear motion
is in addition to the linear motion of the body’s center of mass. To get the total linear
motion of any particle or point on the rigid body, you must be able to relate the angular
motion of the body to the linear motion of the particle or point as it sweeps its circular
path about the axis of rotation.
Before we show you how to do this, we’ll explain why you would even want to perform
such a calculation. Basically, in dynamics, knowing that two objects have collided is not
always enough, and you’ll often want to know how hard, so to speak, these two objects
have collided. When you’re dealing with interacting rigid bodies that may at some point
make contact with one another or with other fixed objects, you need to determine not
only the location of the points of contact, but also the relative velocity or acceleration
between the contact points. This information will allow you to calculate the interaction
forces between the colliding bodies.
The arc length of the path swept by a particle on the rigid body is a function of the
distance from the axis of rotation to the particle and the angular displacement, Cl. We’ll
use c to denote arc length and r to denote the distance from the axis of rotation to the
particle, as shown in Figure 2-9.
Angular Velocity and Acceleration | 63
www.it-ebooks.info
Figure 2-9. Circular path of particles making up a rigid body
The formula relating arc length to angular displacement is:
c = r £2
where Cl must be in radians, not degrees. If you differentiate this formula with respect
to time:
dc/dt = r d£2/dt
you get an equation relating the linear velocity of the particle as it moves along its circular
path to the angular velocity of the rigid body. This equation is written as follows:
v = r co
This velocity as a vector is tangent to the circular path swept by the particle. Imagine
this particle as a ball at the end of a rod where the other end of the rod is fixed to a
rotating axis. If the ball is released from the end of the rod as it rotates, the ball will fly
off in a direction tangent to the circular path it was taking when attached to the rod.
This is in the same direction as the tangential linear velocity given by the preceding
equation. Figure 2-10 illustrates the tangential velocity.
64 | Chapter 2: Kinematics
www.it-ebooks.info
Figure 2-10. Linear velocity due to angular velocity
Differentiating the equation, v = r to:
dv/dt = r dco/dt
yields this formula for the tangential linear acceleration as a function of angular accel¬
eration:
a t = r a
Note that there is another component of acceleration for the particle that results from
the rotation of the rigid body. This component—the centripetal acceleration—is normal,
or perpendicular, to the circular path of the particle and is always directed toward the
axis of rotation. Remember that velocity is a vector and since acceleration is the rate of
change in the velocity vector, there are two ways that acceleration can be produced. One
way is by a change in the magnitude of the velocity vector—that is, a change in speed
—and the other way is a change in the direction of the velocity vector. The change in
speed gives rise to the tangential acceleration component, while the direction change
gives rise to the centripetal acceleration component. The resultant acceleration vector
is the vector sum of the tangential and centripetal accelerations (see Figure 2-11). Cen¬
tripetal acceleration is what you feel when you take your car around a tight curve even
though your speed is constant.
Angular Velocity and Acceleration | 65
www.it-ebooks.info
Figure 2-11. Tangential and centripetal acceleration
The formula for the magnitude of centripetal acceleration, a n , is:
where v is the tangential velocity. Substituting the equation for tangential velocity into
this equation for centripetal acceleration gives the following alternative form:
a n = r co 2
In two dimensions you can easily get away with using these scalar equations; however,
in three dimensions you’ll have to use the vector forms of these equations. Angular
velocity as a vector is parallel with the axis of rotation. In Figure 2-10 the angular velocity
would be pointing out of the page directly at you. Its sense, or direction of rotation, is
determined by the right hand rule. If you curl the fingers of your right hand in an arc
around the axis of rotation with your fingers pointing toward the direction in which
the body is rotating, then your thumb will stick up in the direction of the angular velocity
vector.
If you take the vector cross product (refer to the sidebar “Vector Cross Product” on page
67 for background and to Appendix A for a review of vector math) of the angular velocity
vector and the vector from the axis of rotation to the particle under consideration, you’ll
end up with the linear, tangential velocity vector. This is written as follows:
v = (o X r
66 | Chapter 2: Kinematics
www.it-ebooks.info
Note that this gives both the magnitude and direction of the linear, tangential velocity.
Also, be sure to preserve the order of the vectors when taking the cross product—that
is, to cross r, and not the other way around, which would give the wrong direction for v.
Vector Cross Product
Given any two vectors A and B, the cross product A x B is defined by a third vector C
with a magnitude equal to AB sin 0, where 0 is the angle between the two vectors A and
B, as illustrated in the following figure.
C = A x B
C = AB sin 0
The direction of C is determined by the right hand rule. As noted previously, the right
hand rule is a simple trick to help you keep track of vector directions. Assume that A
and B lie in a plane and let an axis of rotation extend perpendicular to this plane through
a point located at the tail of A. Pretend to curl the fingers of your right hand around the
axis of rotation from vector A toward B. Now extend your thumb (as though you are
giving a thumbs up) while keeping your fingers curled around the axis. The direction
that your thumb is pointing indicates the direction of vector C.
In the preceding figure, a parallelogram is formed by A and B (the shaded region). The
area of this parallelogram is the magnitude of C, which is AB sin 0.
There are two equations that you’ll need in order to determine the vectors for tangential
and centripetal acceleration. They are:
a n = (o X (w X r)
a t = a x r
Another way to look at the quantities v, a n , and a t is that they are the velocity and
acceleration of the particle under consideration, on the rigid body, relative to the point
about which the rigid body is rotating—for example, the body’s center-of-mass location.
This is very convenient because, as we said earlier, you’ll want to track the motion of
the rigid body as a particle when viewing the big picture without having to worry about
Angular Velocity and Acceleration | 67
www.it-ebooks.info
what each particle making up the rigid body is doing all the time. Thus, you treat the
rigid body’s linear motion and its angular motion separately. When you do need to take
a close look at specific particles of—or points on—the rigid body, you can do so by taking
the motion of the rigid body as a particle and then adding to it the relative motion of
the point under consideration.
Figure 2-12 shows a rigid body that is traveling at a speed v cg , where v cg is the speed of
the rigid body’s center of mass (or center of gravity). Remember, the center of mass is
the point to track when treating a rigid body as a particle. This rigid body is also rotating
with an angular velocity to about an axis that passes through the body’s center of mass.
The vector r is the vector from the rigid body’s center of mass to the particular point of
interest, P, located on the rigid body.
In this case, we can find the resultant velocity of the point, P, by taking the vector sum
of the velocity of the body’s center of mass and the tangential velocity of point P due to
the body’s angular velocity to. Here’s what the vector equation looks like:
Vr = v cg + v t
or:
VR = V cg + (to x r)
You can do the same thing with acceleration to determine point P’s resultant accelera¬
tion. Here you’ll take the vector sum of the acceleration of the rigid body’s center of
68 | Chapter 2: Kinematics
www.it-ebooks.info
mass, the tangential acceleration due to the body’s angular acceleration, and the cen¬
tripetal acceleration due to the change in direction of the tangential velocity. In equation
form, this looks like:
a R — a cg + a n + a t
Figure 2-13 illustrates what’s happening here.
You can rewrite the equation for the resultant acceleration in the following form:
a R = a c „ + (w x (to x r)) + (a x r)
As you can see, using these principles of relative velocity and acceleration allows you to
calculate the resultant kinematic properties of any point on a rigid body at any given
time by determining what the body’s center of mass is doing along with how the body
is rotating.
Angular Velocity and Acceleration | 69
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 3
Force
This chapter is a prerequisite to Chapter 4, which addresses the subject of kinetics. The
aim here is to provide you with enough of a background on forces so you can readily
appreciate the subject of kinetics. This chapter is not meant to be the final word on the
subject of force. In fact, we feel that the subject of force is so important to realistic
simulations that we’ll revisit it several times in various contexts throughout the re¬
mainder of this book. In this chapter, we’ll discuss the two fundamental categories of
force and briefly explain some important specific types of force. We’ll also explain the
relationship between force and torque.
Forces
As we mentioned in Chapter 2, you need to understand the concept of force before you
can fully understand the subject of kinetics. Kinematics is only half the battle. You are
already familiar with the concept of force from your daily experiences. You exert a force
on this book as you hold it in your hands, counteracting gravity. You exert force on your
mouse as you move it from one point to another. When you play soccer, you exert force
on the ball as you kick it. In general, force is what makes an object move, or more
precisely, what produces an acceleration that changes the velocity. Even as you hold this
book, although it may not be moving, you’ve effectively produced an acceleration that
cancels the acceleration from gravity. When you kick that soccer ball, you change its
velocity from, say, 0 when the ball is at rest to some positive value as the ball leaves your
foot. These are some examples of externally applied contact forces.
There’s another broad category of forces, in addition to contact forces, called field forces
or sometimes/orces at a distance. These forces can act on a body without actually having
to make contact with it. A good example is the gravitational attraction between objects.
Another example is the electromagnetic attraction between charged particles. The con¬
cept of a force field was developed long ago to help us visualize the interaction between
objects subject to forces at a distance. You can say that an object is subjected to the
71
www.it-ebooks.info
gravitational field of another object. Thinking in terms of force fields can help you grasp
the fact that an object can exert a force on another object without having to physically
touch it.
Within these two broad categories of forces, there are specific types of forces related to
various physical phenomena—forces due to friction, buoyancy, and pressure, among
others. We’ll discuss idealizations of several of these types of forces in this chapter. Later
in this book, we’ll revisit these forces from a more practical point of view.
Before going further, we need to explain the implications of Newton’s third law as in¬
troduced in Chapter 1. Remember, Newton’s third law states that for every force acting
on a body, there is an equal and opposite reacting force. This means that forces must
exist in pairs—a single force can’t exist by itself.
Consider the gravitational attraction between the earth and yourself. The earth is ex¬
erting a force—your weight—on you, accelerating you toward its center. Likewise, you
are exerting a force on the earth, accelerating it toward you. The huge difference between
your mass and the earth’s makes the acceleration of the earth in this case so small that
it’s negligible. Earlier we said you are exerting a force on this book to hold it up; likewise,
this book is exerting a force on your hands equal in magnitude but opposite in direction
to the force you are exerting on it. You feel this reaction force as the book’s weight.
This phenomenon of action-reaction is the basis for rocket propulsion. A rocket engine
exerts force on the fuel molecules that are accelerated out of the engine exhaust nozzle.
The force required to accelerate these molecules is exerted back against the rocket as a
reaction force called thrust. Throughout the remainder of this book, you’ll see many
other examples of action-reaction, which is an important phenomenon in rigid-body
dynamics. It is especially important when we are dealing with collisions and objects in
contact, as you’ll see later.
Force Fields
The best example of a force field or force at a distance is the gravitational attraction
between objects. Newton’s law of gravitation states that the force of attraction between
two masses is directly proportional to the product of the masses and inversely propor¬
tional to the square of the distances separating the centers of each mass. Further, this
law states that the line of action of the force of attraction is along the line that connects
the centers of the two masses. This is written as follows:
F a = (Gm 1 m 2 )/r 2
where G is the gravitational constant, Newton’s so-called universal constant. G was first
measured experimentally by Sir Henry Cavendish in 1798 and equals 6.673x10 11 (N-
m 2 )/kg 2 in metric units or 3.436x10 s ft 4 /(lb-s 4 ) in English units.
72 | Chapter 3:Force
www.it-ebooks.info
So far in this book, I’ve been using the acceleration due to gravity, g, as a constant 9.8
m/s 2 (32.174 ft/s 2 ). This is true when you are near the earths surface—for example, at
sea level. In reality, g varies with altitude—maybe not by much for our purposes, but it
does. Consider Newton’s second law along with the law of gravitation for a body near
the earth. Equating these two laws, in equation form, yields:
m a = (G M e m) / (R e + h ) 2
where m is the mass of the body, a is the acceleration of the body due to the gravitational
attraction between it and the earth, M e is the earth’s mass, R e is the radius of the earth,
and h is the altitude of the body. If you solve this equation for a, you’ll have a formula
for the acceleration due to gravity as a function of altitude:
a = g’ = (G M e ) / (R e + h ) 2
The radius of the earth is approximately 6.38xl0 6 m, and its mass is about 5.98xl0 24
kgs. Substituting these values in the preceding equation and assuming 0 altitude (sea
level) yields the constant g that we’ve been using so far—that is, g at sea level equals 9.8
m/s 2 .
Friction
Frictional forces (friction) always resist motion and are due to the interaction between
contacting surfaces. Thus, friction is a contact force. Friction is always parallel to the
contacting surfaces at the point of contact—that is, friction is tangential to the contacting
surfaces. The magnitude of the frictional force is a function of the normal force between
the contacting surfaces and the surface roughness.
This is easiest to visualize by looking at a simple block on a horizontal surface, as shown
in Figure 3-1.
Friction | 73
www.it-ebooks.info
In Figure 3-1, the block is resting on the horizontal surface with a small force, F a , applied
to the block on a line of action through the blocks center of mass. As this applied force
increases, a frictional force will develop between the block and the horizontal surface,
tending to resist the motion of the block. The maximum value of this frictional force is:
Ffmax = Ms N
where p s is the experimentally determined coefficient of static 1 friction and N is the
normal (perpendicular) force between the block and the surface, which equals the
weight of the block in this case. As the applied force increases but is still less than F [max ,
the block will remain static and F { will be equal in magnitude to the applied force. The
block is in static equilibrium. When the applied force becomes greater than F' fmax , the
frictional force can no longer impede the block’s motion and the block will accelerate
under the influence of the applied force. Immediately after the block starts its motion,
the frictional force will decrease from F fmax to F fk , where is:
Ffk = Pk N
Here k means kinetic since the block is in motion, and p k , the coefficient of kinetic
friction, 2 is less than p s . Like the static coefficient of friction, the kinetic coefficient of
friction is determined experimentally. Table 3 -1 shows typical coefficients of friction for
several surfaces in contact.
Table 3-1. Coefficients of friction of common surfaces
1 Surface condition
M u
% difference I
Dry glass on glass
0.94
0.4
54%
Dry iron on iron
1.1
0.15
86%
Dry rubber on pavement
0.55
0.4
27%
Dry steel on steel
0.78
0.42
46%
Dry Teflon on Teflon
0.04
0.04
—
Dry wood on wood
0.38
0.2
47%
Ice on ice
0.1
0.03
70%
Oiled steel on steel
0.10
0.08
20%
The data in Table 3 -1 is provided here to show you the magnitude of some typical friction
coefficients and the relative difference between the static and kinetic coefficients for
certain surface conditions. Other data is available for these and other surface conditions
1. Static here implies that there is no motion; the block is sitting still with all forces balancing.
2. The term dynamic is sometimes used here instead of kinetic.
74 | Chapter 3: Force
www.it-ebooks.info
in the technical literature (see the Bibliography for sources). Note that experimentally
determined friction coefficient data will vary, even for the same surface conditions,
depending on the specific condition of the material used in the experiments and the
execution of the experiment itself.
Fluid Dynamic Drag
Fluid dynamic drag forces oppose motion like friction. In fact, a major component of
fluid dynamic drag is friction that results from the relative flow of the fluid over (and
in contact with) the body’s surface. Friction is not the only component of fluid dynamic
drag, though. Depending on the shape of the body, its speed, and the nature of the fluid,
fluid dynamic drag will have additional components due to pressure variations in the
fluid as it flows around the body. If the body is located at the interface between two fluids
(like a ship on the ocean where the two fluids are air and water), an additional compo¬
nent of drag will exist due to the wave generation.
In general, fluid dynamic drag is a complicated phenomenon that is a function of several
factors. We won’t go into detail in this section on all these factors, since we’ll revisit this
subject later. However, we do want to discuss how the viscous (frictional) component of
these drag forces is typically idealized.
Ideal viscous drag is a function of velocity and some experimentally determined drag
coefficient that’s supposed to take into account the surface conditions of the body, the
fluid properties (density and viscosity), and the flow conditions. You’ll typically see a
formula for viscous drag force in the form:
F v — -C f v
where C f is the drag coefficient, v is the body’s speed, and the minus sign means that the
force opposes motion. This formula is valid for slow-moving objects in a viscous fluid.
“Slow moving” implies that the flow around the body is laminar, which means that the
flow streamlines are undisturbed and parallel.
For fast-moving objects, you’ll use the formula for F v written as a function of speed
squared as follows:
F v = -C f v 2
“Fast moving” implies that the flow around the object is turbulent, which means that
the flow streamlines are no longer parallel and there is a sort of mixing effect in the flow
around the object. Note that the values of C f are generally not the same for these two
equations. In addition to the factors mentioned earlier, C f depends significantly on
whether the flow is laminar or turbulent.
Fluid Dynamic Drag | 75
www.it-ebooks.info
Both of these equations are very simplified and inadequate for practical analysis of fluid
flow problems. However, they do offer certain advantages in computer game simula¬
tions. Most obviously, these formulas are easy to implement—you need only know the
velocity of the body under consideration, which you get from your kinematic equations,
and an assumed value for the drag coefficient. This is convenient, as your game world
will typically have many different types of objects of all sizes and shapes that would make
rigorous analysis of each of their drag properties impractical. If the illusion of realism
is all you need, not real-life accuracy, then these formulas may be sufficient.
Another advantage of using these idealized formulas is that you can tweak the drag
coefficients as you see fit to help reduce numerical instabilities when solving the equa¬
tions of motion, while maintaining the illusion of realistic behavior. If real-life accuracy
is what you’re going for, then you’ll have no choice but to consider a more involved
(read: complicated) approach for determining fluid dynamic drag. We’ll talk more about
drag in Chapter 6 through Chapter 10.
Pressure
Many people confuse pressure with force. You have probably heard people say, when
explaining a phenomenon, something like, “It pushed with a force of 100 pounds per
square inch.” While you understand what they mean, they are technically referring to
pressure, not force. Pressure is force per unit area, thus the units pounds per square
inch (psi) or pounds per square foot (psf) and so on. Given the pressure, you’ll need to
know the total area acted on by this pressure in order to determine the resultant force.
Force equals pressure times area:
F = PA
This formula tells you that for constant pressure, the greater the area acted upon, the
greater the resultant force. If you rearrange this equation solving for pressure, you’ll see
that pressure is inversely proportional to area—that is, the greater the area for a given
applied force, the smaller the resulting pressure and vice versa.
P = F/A
An important characteristic of pressure is that it always acts normally (perpendicularly)
to the surface of the body or object it is acting on. This fact gives you a clue as to the
direction of the resultant force vector.
We wanted to mention pressure here because you’ll be working with it to calculate forces
when you get to the chapters in this book that cover the mechanics of ships, boats, and
hovercraft. There, the pressures that you’ll consider are hydrostatic pressure (buoyancy)
and aerostatic lift. We’ll take a brief look at buoyancy next.
76 | Chapter 3:Force
www.it-ebooks.info
Buoyancy
You’ve no doubt felt the effects of buoyancy when immersing yourself in the bathtub.
Buoyancy is why you feel lighter in water than you do in air and why some people can
float on their backs in a swimming pool.
Buoyancy is a force that develops when an object is immersed in a fluid. It’s a function
of the volume of the object and the density of the fluid and results from the pressure
differential between the fluid just above the object and the fluid just below the object.
Pressure increases the deeper you go in a fluid, thus the pressure is greater at the bottom
of an object of a given height than it is at the top of the object. Consider the cube shown
in Figure 3-2.
n n
u n
Figure 3-2. Immersed cube
Let s denote the cube’s length, width, and height, which are all equal. Further, let h t
denote the depth to the top of the cube and h b the depth to the bottom of the cube. The
pressure at the top of the cube is P t = p g h t , which acts over the entire surface area of
the top of the cube, normal to the surface in the downward direction. The pressure at
the bottom of the cube is P b = p g h b , which acts over the entire surface area of the bottom
of the cube, normal to the surface in the upward direction. Note that the pressure acting
on the sides of the cube increases linearly with submergence, from P t to P b . Also, note
that since the side pressure is symmetric, equal and opposite, the net side pressure is 0,
which means that the net side force (due to pressure) is also 0. The same is not true of
the top and bottom pressures, which are obviously not equal, although they are opposite.
The force acting down on the top of the cube is equal to the pressure at the top of the
cube times the surface area of the top. This can be written as follows:
Ft = Pt A t
Buoyancy | 77
www.it-ebooks.info
F t = (p g h t ) (s 2 )
Similarly, the force acting upward on the bottom of the cube is equal to the pressure at
the bottom times the surface area of the bottom.
Fb = Pb A b
Fb = (P g h b ) (s 2 )
The net vertical force (buoyancy) equals the difference between the top and bottom
forces:
Fb = Fb - F t
Fb = (P g hb) (s 2 ) - (p g hj) (s 2 )
Fb = (P g) (s 2 ) (hb - ht)
This formula gives the magnitude of the buoyancy force. Its direction is straight up,
counteracting the weight of the object.
There is an important observation we need to make here. Notice that ( h b - h t ) is simply
the height of the cube, which is s in this case. Substituting s in place of ( h b - h t ) reveals
that the buoyancy force is a function of the volume of the cube.
F B — (p g) (s 3 )
This is great since it means that all you need to do in order to calculate buoyancy is to
first calculate the volume of the object and then multiply that volume by the specific
weight 3 (p g) of the fluid. In truth, that’s a little easier said than done for all but the
simplest geometries. If you’re dealing with spheres, cubes, cylinders, and the like, then
calculating volume is easy. However, if you’re dealing with any arbitrary geometry, then
the volume calculation becomes more difficult. There are two ways to handle this dif¬
ficulty. The first way is to simply divide the object into a number of smaller objects of
simpler geometry, calculate their volumes, and then add them all up. The second way
is to use numerical integration techniques to calculate volume by integrating over the
surface of the object.
You should also note that buoyancy is a function of fluid density, and you don’t have to
be in a fluid as dense as water to experience the force of buoyancy. In fact, there are
buoyant forces acting on you right now, although they are very small, due to the fact
that you are immersed in air. Water is many times more dense than air, which is why
you notice the force of buoyancy when in water and not in air. Keep in mind, though,
3. Specific weight is density times the acceleration due to gravity. Typical units are lbs/ft 3 and N/m 3 .
78 | Chapter 3:Force
www.it-ebooks.info
that for very light objects with relatively large volumes, the buoyant forces in air may
be significant. For example, consider simulating a large balloon.
Springs and Dampers
Springs are structural elements that, when connected between two objects, apply equal
and opposite forces to each object. This spring force follows Hooke s law and is a func¬
tion of the stretched or compressed length of the spring relative to the rest length of the
spring and the spring constant of the spring. Hooke’s law states that the amount of stretch
or compression is directly proportional to the force being applied. The spring constant
is a quantity that relates the force exerted by the spring to its deflection:
F s = k s (L-r)
Here, F s is the spring force, k s is the spring constant, L is the stretched or compressed
length of the spring, and r is the rest length of the spring. In the metric system of units,
F s would be measured in newtons (1 N = 1 kg-m/s 2 ), L and r in meters, and k s in kg/s 2 .
If the spring is connected between two objects, it exerts a force of F s on one object and
-F s on the other; these are equal and opposite forces.
Dampers are usually used in conjunction with springs in numerical simulations. They
act like viscous drag in that they act against velocity. In this case, if the damper is con¬
nected between two obj ects that are moving toward or away from each other, the damper
acts to slow the relative velocity between the two objects. The force developed by a
damper is proportional to the relative velocity of the connected objects and a damping
constant, k d , that relates relative velocity to damping force.
F d = k d (vi - v 2 )
This equation shows the damping force, F d , as a function of the damping constant and
the relative velocity of the connected points on the two connected bodies. In metric
units, where the damping force is measured in newtons and velocity in m/s, k d has units
of kg/s.
Typically, springs and dampers are combined into a single spring-damper element,
where a single formula represents the combined force. Using vector notation, we can
write the formula for a spring-damper element connecting two bodies as follows:
Fi = -{k s (L - r) + k d (( Vl - v 2 ) • L)/L} L/L
Here, is the force exerted on body 1, while the force, F 2 , exerted on body 2 is:
F 2 = -Ft
Springs and Dampers | 79
www.it-ebooks.info
L is the length of the spring-damper ( L , not in bold print, is the magnitude of the vector
L), which is equal to the vector difference in position between the connected points on
bodies 1 and 2. If the connected objects are particles, then L is equal to the position of
body 1 minus the position of body 2. Similarly, v : and v 2 are the velocities of the con¬
nected points on bodies 1 and 2. The quantity (v, - v 2 ) represents the relative velocity
between the connected bodies.
Springs and dampers are useful when you want to simulate collections of connected
particles or rigid bodies. The spring force provides the structure, or glue, that holds the
bodies together (or keeps them separated by a certain distance), while the damper helps
smooth out the motion between the connected bodies so it’s not too jerky or springy.
These dampers are also very important from a numerical stability point of view in that
they help keep your simulations from blowing up. We’re getting a little ahead of ourselves
here, but we’ll show you how to use these spring-dampers in real-time simulations in
Chapter 13.
Force and Torque
We need to make the distinction here between force and torque. 4 Force is what causes
linear acceleration, while torque is what causes rotational acceleration. Torque is force
times distance. Specifically, to calculate the torque applied by a force acting on an object,
you need to calculate the perpendicular distance from the axis of rotation to the line of
action of the force and then multiply this distance by the magnitude of the force.
This calculation gives the magnitude of the torque. Typical units for force are pounds,
newtons, and tons. Since torque is force times a distance, its units take the form of a
length unit times a force unit (e.g., foot-pounds, newton-meters, or foot-tons).
Since both force and torque are vector quantities, you must also determine the direction
of the torque vector. The force vector is easy to visualize: its line of action passes through
the point of application of the force, with its direction determined by the direction in
which the force is applied. As a vector, the torque’s line of action is along the axis of
rotation, with the direction determined by the direction of rotation and the right hand
rule (see Figure 3-3). As noted in Chapter 2, the right hand rule is a simple trick to help
you keep track of vector directions—in this case, the torque vector. Pretend to curl the
fingers of your right hand around the axis of rotation with your fingertips pointing in
the direction of rotation. Now extend your thumb, as though you are giving a thumbs
up, while keeping your fingers curled around the axis. The direction that your thumb
is pointing indicates the direction of the torque vector. Note that this makes the torque
vector perpendicular to the applied force vector, as shown in Figure 3-3.
4. Another common term for torque is moment.
80 | Chapter 3:Force
www.it-ebooks.info
Figure 3-3. Force and torque
We said earlier that you find the magnitude of torque by multiplying the magnitude of
the applied force times the perpendicular distance between the axis of rotation and the
line of action of the force. This calculation is easy to perform in two dimensions where
the perpendicular distance (d in Figure 3-3) is readily calculable.
However, in three dimensions you’ll want to be able to calculate torque by knowing only
the force vector and the coordinates of its point of application on the body relative to
the axis of rotation. You can accomplish this by using the following formula:
M = r x F
The torque, M, is the vector cross product of the position vector, r, and the force vector,
F.
In rectangular coordinates you can write the distance, force, and torque vectors as fol¬
lows:
r = xi + yj + zk
F = F x i + F y j + F z k
M = M x i + M y j + M z k
The scalar components of r (x, y, and z) are the coordinate distances from the axis of
rotation to the point of application of the force, F. The scalar components of the torque
vector, M, are defined by the following:
M x = y F z - z F y
My = z F x - x F z
Force and Torque | 81
www.it-ebooks.info
M z = x Fy - y F x
Consider the rigid body shown in Figure 3-4 acted upon by the force F at a point away
from the body’s center of mass.
Figure 3-4. Torque example
In this example F, a, and b are given and are as follows:
F = (-90 lbs) i + (156 lbs) j + (0) k
a = 0.66 ft
b = 0.525 ft
Calculate the torque about the body’s center of mass due to the force F.
The first step is to put together the distance vector from the point of application of F to
the body’s center of mass. Since the local coordinates a and b are given, r is simply:
r = (0.66 ft) i + (0.525 ft) j + (0) k
Now using the formula M = r x F (or the formulas for the components of the torque
vector shown earlier), you can write:
M = [(0.66 ft) i + (0.525 ft) j + (0) k] x [(-90 lbs) i + (156 lbs) j
+ (0) k]
M = [(0.66 ft) (156 lbs) - (0.525 ft) (-90 lbs)] k
M = (150.2 ft-lbs) k
82 | Chapter 3:Force
www.it-ebooks.info
Note that the x and y components of the torque vector are 0; thus, the torque moment
is pointing directly along the z-axis. The torque vector would be pointing out of the
page of this book in this case.
In dynamics you need to consider the sum, or total, of all forces acting on an object
separately from the sum of all torques acting on a body. When summing forces, you
simply add, vectorally, all of the forces without regard to their point of application.
However, when summing torques you must take into account the point of application
of the forces to calculate the torques, as shown in the previous example. Then you can
take the vector sum of all torques acting on the body.
When you are considering rigid bodies that are not physically constrained to rotate
about a fixed axis, any force acting through the body’s center of mass will not produce
a torque on the body about its center of gravity. In this case, the axis of rotation passes
through the center of mass of the body and the vector r would be 0 (all components 0).
When a force acts through a point on the body some distance away from its center of
mass, a torque on the body will develop, and the angular motion of the body will be
affected. Generally, field forces, which are forces at a distance, are assumed to act
through a body’s center of mass; thus, only the body’s linear motion will be affected
unless the body is constrained to rotate about a fixed point. Other contact forces, how¬
ever, generally do not act through a body’s center of mass (they could but aren’t neces¬
sarily assumed to) and tend to affect the body’s angular motion as well as its linear
motion.
Summary
As we said earlier, this chapter on forces is your bridge from kinematics to kinetics. Here
you’ve looked at the major force categories—contact forces and force fields—and some
important specific types of forces. This chapter was meant to give you enough theoretical
background on forces so you can fully appreciate the subject of kinetics that’s covered
in the next chapter. In Chapter 15 through Chapter 19, you’ll revisit the subject of forces
from a much more practical point of view when we investigate specific real-life prob¬
lems. We’ll also introduce some new specific types of force in those chapters that we
didn’t cover here.
Summary | 83
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 4
Kinetics
Recall that kinetics is the study of the motion of bodies, including the forces that act on
them. It’s now time that we combine the material presented in the earlier chapters—
namely, kinematics and forces—to study the subject of kinetics. As in Chapter 2 on
kinematics, we’ll first discuss particle kinetics and then go on to discuss rigid-body
kinetics.
In kinetics, the most important equation that you must consider is Newton’s second law:
F = ma
When rigid bodies are involved, you must also consider that the forces acting on the
body will tend to cause rotation of the body in addition to translation. The basic rela¬
tionship here is:
M cg = I a
where M cg is the vector sum of all moments (torques) acting on the body, I is the body
moment of inertia tensor, and a is the angular acceleration.
Collectively, these two equations are referred to as the equations of motion.
There are two types of problems that you will encounter in kinetics. One type is where
you know the force(s) acting on the body, or you can estimate them, and you must solve
for the resulting acceleration of the body (and subsequently its velocity and displace¬
ment) . Another type is where you know the body’s acceleration, or can readily determine
it using kinematics, and you must solve for the force(s) acting on the body.
85
www.it-ebooks.info
This chapter will primarily discuss the first type of problem, where you know the force(s)
acting on the body, which is more common to in-game physics. The second type of
problem has become important with the advent of motion-based controllers such as the
Sony SixAxis and Nintendo Wii Remote. These controllers rely on digital accelerome¬
ters to directly measure the acceleration of a controller. While this is most often used to
find the controller’s orientation, it is also possible to integrate the time history of these
sensor values to determine velocity and position. Additionally, if you know the mass of
the controller or device, you can find the force. Accelerometers are found in most
smartphones as well, which also allows for the use of kinematic-based input. So as to
not confuse the two types of problems, we’ll discuss the second type, with the acceler¬
ation as input, in detail in Chapter 21.
Let us stress that you must consider the sum of all of the forces acting on the body when
solving kinetics problems. These include all applied forces and all reaction forces. Aside
from the computational difficulties of solving the equations of motion, one of the more
challenging aspects of kinetics is identifying and properly accounting for all of these
forces. In later chapters, you’ll look at specific problems where we’ll investigate the
particular forces involved. For now, and for the purpose of generality, let’s stick with the
idealized forces introduced in the previous chapter.
Here is the general procedure for solving kinetics problems of interest to us:
1. Calculate the body’s mass properties (mass, center of mass, and moment of inertia).
2. Identify and quantify all forces and moments acting on the body.
3. Take the vector sum of all forces and moments.
4. Solve the equations of motion for linear and angular accelerations.
5. Integrate with respect to time to find linear and angular velocity.
6. Integrate again with respect to time to find linear and angular displacement.
This outline makes the solution to kinetics problems seem easier than it actually is
because there are a number of complicating factors that you’ll have to overcome. For
example, in many cases the forces acting on a body are functions of displacement, ve¬
locity, or acceleration. This means that you’ll have to use iterative techniques in order
to solve the equations of motion. Further, since you most likely will not be able to derive
closed-form solutions for acceleration, you’ll have to numerically integrate in order to
estimate velocity and displacement at each instant of time under consideration. These
computational aspects will be addressed further in Chapter 7 through Chapter 13.
86 | Chapter 4: Kinetics
www.it-ebooks.info
Particle Kinetics in 2D
As in particle kinematics, in particle kinetics you need to consider only the linear motion
of the particle. Thus, the equations of motion will consist of equations of the form F =
ma, where motion in each coordinate direction will have its own equation. The equa¬
tions for 2D particle motion are:
XF x = m a x
XF y = m a y
where XF x means the sum of all forces in the x-direction, XF y means the sum of all forces
in the y-direction, a x is the acceleration in the x-direction, and a y is the acceleration in
the y-direction.
The resultant force and acceleration vectors are:
a = a x i + a y j
a — + a y
XF = XF x i + XF y j
IF = ^(ZF,f + (ZF,f
Let’s look at an example that appears simple but demonstrates the complexity of finding
closed-form solutions. A ship floating in water, initially at rest, starts up its propeller
generating a thrust, T, which starts the ship moving forward. Assume that the ship’s
forward speed is slow and the resistance to its motion can be approximated by:
R = -C v
where R is the total resistance, C is a drag coefficient, v is the ship speed, and the minus
sign indicates that this resistive force opposes the forward motion of the ship. Find
formulas for the ship’s speed, acceleration, and distance traveled as functions of time,
assuming that the propeller thrust and resistance force vectors act on a line of action
passing through the ship’s center of gravity. This assumption lets you treat the ship as a
particle instead of a rigid body.
The first step in solving this problem is to identify all of the forces acting on the ship.
Figure 4-1 shows a free-body diagram of the ship with all of the forces acting on it—
namely, the propeller thrust, T; resistance, R; the ship’s weight, W; and buoyancy, B.
Particle Kinetics in 2D | 87
www.it-ebooks.info
J
i
f
B
T -► <
>-x < - R
1
1/
Figure 4-1. Free-body diagram of ship
Notice here that the buoyancy force is exactly equal in magnitude to the ship’s weight
and opposite in direction; thus, these forces cancel each other out and there will be no
motion in the y-direction. This must be the case if the ship is to stay afloat. This obser¬
vation effectively reduces the problem to a one-dimensional problem with motion in
the x-direction, only where the forces acting in the x-direction are the propeller thrust
and resistance.
Now you can write the equation (for motion in the x-direction) using Newton’s second
law, as follows:
£F = m a
T - R = m a
T - (C v) = m a
Where a is the acceleration in the x-direction, and v is the speed in the x-direction.
The next step is to integrate this equation of motion in order to derive a formula for the
speed of the ship as a function of time. To do this, you must make the substitution a =
dv/dt, rearrange, integrate, and then solve for speed as follows:
T - (C v) = m (dv/dt)
dt = (m / (T-Cv)) dv
/(0 to t) dt = /(v i to v2) (m / (T-Cv)) dv
t - 0 = -(m/C) ln(T-Cv) I (vl t0 v2)
t = -(m/C) ln(T-Cv 2 ) + (m/C) ln(T-Cvi)
t = (m/C) [ln(T-Cv!) - ln(T-Cv 2 )]
(C/m) t = In [(T-C Vl )/(T-Cv 2 )]
e (C/m) t _ e In [(T-Cvl) / (T-Cv2)]
e (C/m) t = (T-CvO / (T-Cv 2 )
(T-Cv 2 ) = (T-Cvi)
88 | Chapter 4: Kinetics
www.it-ebooks.info
V 2 = (T/C) - e-( c/m > 1 (T/C - vi)
where v 1 is the initial ship speed (which is constant) and v 2 is the ship speed at time t.
v 2 is what you’re after here, since it tells you how fast the ship is traveling at any instant
of time.
Now that you have an equation for speed as a function of time, you can derive an
equation for displacement (distance traveled, in this case) as a function of time. Here,
you’ll have to recall the formula v dt = ds, substitute the previous formula for speed,
integrate, rearrange, and solve for distance traveled. These steps are shown here:
v dt = ds
v 2 dt = ds
((T/C) - e-( c/m > 1 (T/C - vO) dt = ds
/(0 to t) (T/C) - e-( c/m > 1 (T/C - Vl ) dt = / (sl t0 s2) ds
(T/C) / ( o to t) dt - (T/C - Vl ) / ( o to t) e- (C/m) 1 dt = s 2 - Sl
[(T/C) t + ((T/C) - vtXm/C) e-( c/m > l ] (010 1 ) = s 2 - si
[(T/C) t + ((T/C) - vi)(m/C) e-( c/m ) l ] - [0 + ((T/C) - vi)(m/C)] =
S2-N
(T/C) t + (T/C - vi) (m/C) e-( c/m > 1 - (T/C - vi) (m/C) = s 2 -s,
s 2 = Sl + (T/C) t + (T/C - vO (m/C) e -( c/m > 1 - (T/C - vi) (m/C)
Finally you can write an equation for acceleration by going back to the original equation
of motion and solving for acceleration:
T - (C v) = m a
a = (T - (C v)) / m
where:
v = v 2 = (T/C) - e -( c/m ) 1 (T/C - vi)
In summary, the equations for velocity, distance traveled, and acceleration are as follows:
v 2 = (T/C) - e-( c/m > 1 (T/C - vi)
s 2 = Sj + (T/C) t + (T/C - vi) (m/C) e -( c/m ) 1 - (T/C - vi) (m/C)
a = (T - (C v)) / m
To illustrate the motion of the ship further, we’ve plotted the ship’s speed, distance trav¬
eled, and acceleration versus time, as shown in Figure 4-2, Figure 4-3, and Figure 4-4.
To facilitate these illustrations, we’ve assumed the following:
Particle Kinetics in 2D | 89
www.it-ebooks.info
• The initial ship speed and displacement are 0 at time 0.
• The propeller thrust is 20,000 thrust units.
• The ship’s mass is 10,000 mass units.
• The drag coefficient is 1,000.
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43
Time
Figure 4-2. Speed versus time
Time
Figure 4-3. Distance versus time
90 | Chapter 4: Kinetics
www.it-ebooks.info
1 4 7 10 13 16 19 22 25 28 31 34 37 40 43
Time
Figure 4-4. Acceleration versus time
You’ll notice that the ship’s speed approaches the steady state speed of 20 speed units,
assuming that the propeller thrust remains constant. This corresponds to a reduction
in acceleration from a maximum acceleration at time 0 to no acceleration once the steady
speed is achieved.
This example illustrates how to set up the differential equations of motion and integrate
them to find velocity, displacement, and acceleration. In this case, you were able to find
a closed-form solution—that is, you were able to integrate the equations symbolically
to derive new ones. You could do this because we imposed enough constraints on the
problem to make it manageable. But you can readily see that if there were more forces
acting on the ship, or if the thrust were not held constant but was some function of
speed, or if the resistance were a function of speed squared, and so on, the problem gets
increasingly complicated—making a closed-form solution much more difficult, if not
impossible.
Particle Kinetics in 3D
As in kinematics, extending the equations of motion for a particle to three dimensions
is easy to do. You simply need to add one more component and will end up with three
equations as follows:
SF X = m a x
SF y = m a y
Particle Kinetics in 3D | 91
www.it-ebooks.info
XF Z = m a z
The resultant force and acceleration vectors are now:
a = a x i + a y j + a z k
a = + ciy + dj
XF = XF x i + XFy j + XF Z k
zf = ^(SF I f + (SF y f + (SF.f
To hammer these concepts home, we want to present another example.
Let’s go back to the cannon example program discussed in Chapter 2. In that example,
we made some simplifying assumptions so we could focus on the kinematics of the
problem without complicating it too much. One of the more significant assumptions
we made was that there was no drag acting on the projectile as it flew through the air.
Physically, this would be valid only if the projectile were moving through a vacuum,
which, of course, is unlikely here on Earth. Another significant assumption we made
was that there was no wind to act on the projectile and affect its course. These two
considerations, drag and wind, are important in real-life projectile problems, so to make
this example a little more interesting—and more challenging to the user if this were an
actual game—we’ll add these two considerations now.
First, assume that the projectile is a sphere and the drag force acting on it as it flies
through the air is a function of some drag coefficient and the speed of the projectile.
This drag force can be written as follows:
F d = -C d v
Fd = — C d v x i - C d Vy j — c-d v z k
where C d is the drag coefficient, v is the velocity of the projectile (v x , v y , and v z are its
components), and the minus sign means that this drag force opposes the projectile’s
motion. Actually, we’re cheating a bit here since in reality the fluid dynamic drag would
be more a function of speed squared. We’re doing this here to facilitate a closed-form
solution. Also, the drag coefficient here would be determined experimentally for each
shape. Later, we’ll discuss how experimental data is used on basic shapes to give data for
similar ships.
Second, assume that the projectile is subjected to a blowing wind, the force of which is
a function of some drag coefficient and the wind speed. This force can be written as
follows:
F w — C w v w
F w = — C w v wx 1 — C w v wz k
92 | Chapter 4: Kinetics
www.it-ebooks.info
where C w is the drag coefficient, v w is the wind speed, and the minus sign means that
this force opposes the projectiles motion when the wind is blowing in a direction op¬
posite of the projectile’s direction of motion. When the wind is blowing with the pro¬
jectile—say, from behind it—then the wind will actually help the projectile along instead
of impede its motion. In general, C w is not necessarily equal to the C d shown in the drag
formula. Referring to Figure 2-3, we’ll define the wind direction as measured by the
angle y. The x and z components of the wind force can now be written in terms of the
wind direction, y, as follows:
F\vx = F w cos y = —(C w v w ) cos y
Fwz = F w sin y = -(C w v w ) sin y
We ignored the y-direction as we assume the wind is flowing parallel to the ground.
Finally, let’s apply a gravitational force to the projectile instead of specifying the effect
of gravity as a constant acceleration, as we did in Chapter 2. This allows you to include
the force due to gravity in the equations of motion. Assuming that the projectile is
relatively close to sea level, the gravitational force can be written as:
F g = -m g j
where the minus sign indicates that it acts in the negative y-direction (pulling the pro¬
jectile toward the earth), and yon the righthand side of this equation is the acceleration
due to gravity at sea level.
Now that all of the forces have been identified, you can write the equations of motion
in each coordinate direction:
£F X = F wx + F dx = m (dv x /dt)
SFy = F dy + F gy = m (dvy/dt)
SF Z = F wz + F dz = m (dv z /dt)
Note here that we already made the substitution dv/dt for acceleration in each equation.
Following the same procedure shown in the previous section, you now need to integrate
each equation of motion twice—once to find an equation for velocity as a function of
time, and another to find an equation for displacement as a function of time. As before,
we’ll show you how this is done component by component.
You might be asking yourself, where’s the thrust force from the cannon that propels the
projectile in the first place? In this example, we’re looking specifically at the motion of
the projectile after it has left the muzzle of the cannon, where there is no longer a thrust
force acting on the projectile; it isn’t self-propelled. To account for the effect of the
cannon thrust force, which acts over a very short period of time while the projectile is
within the cannon, you have to consider the muzzle velocity of the projectile when it
initially leaves the cannon. The components of the muzzle velocity in the coordinate
Particle Kinetics in 3D | 93
www.it-ebooks.info
directions will become initial velocities in each direction, and they will be included in
the equations of motion once they’ve been integrated. The initial velocities will show
up in the velocity and displacement equations just like they did in the example in
Chapter 2. You’ll see this in the following sections.
X Components
The first step is to make the appropriate substitutions for the force terms in the equation
of motion, and then integrate to find an equation for velocity.
-F wx - Fdx = m(dv x /dt)
-(C w v w cos y) - C d v x = m dv x /dt
dt = m dv x / [-(C w v w cos y) - C d v x ]
/(0 to t) dt = /(vx 1 to vx2) —m / [(C w v w cos y) + C d v x ] dv x
t = — (m/C d ) ln((C w v w cos y) + C d v x )l ( vx j to vx2)
t = -(m/C d ) ln((C w v w cos y) + C d v x2 ) + (m/C d ) ln((C w v w cos y)
+ Cd v x i)
(C d /m) t = ln[((C w v w cos y) + C d v x! ) / ((C w v w cos y) + C d v x2 )]
P (C , /m) t _ P ln[((C v cos y) + Cd v ) / ((Cw vw cos y) + C , v .)]
c d —c ww xl dx2
e (Cd/m) t = ( '( Cw Vw cos + Cd Vxl ) / ((C w V w cos y) + C d v x2 )
((C w v w cos y) + C d v x2 ) = ((C w v w cos y) + C d v xl ) e-( c d /m) 1
v x2 = (1/C d ) [ eC c d /m > 1 (c w v w cos y + C d v xl ) - (C w v w cos y)]
To get an equation for displacement as a function of time, you need to recall the equation
v dt= ds, make the substitution for v (using the preceding equation) and then integrate
one more time.
v x2 dt = ds x
(1/C d ) [et -C d /m ) 1 (c w v w cos y + C d v x i) - (c w v w cos y)] dt = ds x
/(0 to t) (1/Cd) [e ( “ c d /m > 1 (c w v w cos y + C d v x! ) - (c w v w cos y)] dt
= /sxl to sx2) ds x
s x 2 = [(m/C d ) el- 0 /™! 1 (-(C w v w cos y) / C d - v xl ) - ((C w v w cos
y)/c d ) t]-
[(m/C d ) (-(C w v w cos y) / C d - v xl )] + s xl
Yes, these equations are ugly. Just imagine if we hadn’t made the simplifying assumption
that drag is proportional to speed and not speed squared! You would have ended up
with some really nice equations with an arctan term or two thrown in.
94 | Chapter 4: Kinetics
www.it-ebooks.info
Y Components
For the y components, you need to follow the same procedure shown earlier for the x
components, but with the appropriate y-direction forces. Here’s what it looks like:
-F dy - F gy = m (dvy/dt)
—C d v y - m g = m (dvy/dt)
f (0 to t) dt = —m /(vyi to vy2) l/(C d v y + m g) dvy
v y2 = (l/c d ) (C d v yl + m g) - (m g)/C d
Now that you have an equation for velocity, you can proceed to get an equation for
displacement as before:
v y 2 dt = ds y
[(1/C d ) e(- c d /m ) 1 (Q v yl + m g) - (m g)/C d ] dt = ds y
/(0 to t) t(l/Cd) e ( -- c d /m - >t (C d vyi + m g) - (m g)/C d ] dt = / (sy i t0
sy2) dSy
Sy2 = Syl + [ (v y i + (m g)/Cd) (m/C d ) e(- c d /m)t - t (m g)/C d ] +
[(m/C d )(v y i + (m g)/C d )]
OK, that’s two down and only one more to go.
Z Components
With the z component, you get a break. You’ll notice that the equations of motion for
the x and z components look almost the same with the exception of the x and z subscripts
and the sine versus cosine terms. Taking advantage of this fact, you can simply copy the
x component equations and replace the x subscript with a z and the cosine terms with
sines and be done with it:
v z2 = (1/C d ) [e ( - c d /m)t (c w v w sin y + C d v z! ) - (c w v w sin y)]
s z2 = [(m/C d ) e(- c d /m)t (-(C w v w sin y) / C d - v z i) - ((C w v w sin
y)/c d ) t] -
[(m/C d ) (-(C w v w sin y)/C d - v z i)] + s z!
Cannon Revised
Now that you have some new equations for the projectile’s displacement in each coor¬
dinate direction, you can go to the cannon example source code and replace the old
displacement calculation formulas with the new ones. Make the changes in the DoSimu
lation function as follows:
Particle Kinetics in 3D | 95
www.it-ebooks.info
//.-//
Int DoSinulation(void)
//.-//
{
// new local variables:
double sxl, vxl;
double syl, vyl;
double szl, vzl;
// Now we can calculate the position vector at this tine
// Old position vector connented out:
//s.i = Vn * cosX * tine + xe;
//s.j = (Yb + L * cos(Alpha*3.14/180)) + (Vn * cosY * tine) -
(0.5 * g * tine * tine);
//s.k = Vn * cosZ * tine + ze;
// New position vector calculations:
sxl = xe;
vxl = Vn * cosX;
syl = Yb + L * cos(Alpha * 3.14/180);
vyl = Vn * cosY;
szl = ze;
vzl = Vn * cosZ;
s.i =((n/Cd) * exp(-(Cd * tine)/n) * ((-Cw * Vw * cos(GannaW * 3.14/180))/Cd -
vxl) - (Cw * Vw * cos(GannaW * 3.14/180) * tine) / Cd ) -
( (n/Cd) * ((-Cw * Vw * cos(GannaW * 3.14/180))/Cd - vxl) ) + sxl;
s.j = syl + ( -(vyl + (n * g)/Cd) * (n/Cd) * exp(-(Cd*tine)/n) -
(n * g * tine) / Cd ) + ( (n/Cd) * (vyl + (n * g)/Cd) );
s.k =((n/Cd) * exp(-(Cd * tine)/n) * ((-Cw * Vw * sin(GannaW * 3.14/180))/Cd -
vzl) - (Cw * Vw * sin(GannaW * 3.14/180) * tine) / Cd ) -
( (n/Cd) * ((-Cw * Vw * sin(GannaW * 3.14/180))/Cd - vzl) ) + szl;
}
To take into account the cross wind and drag, you’ll need to add some new global
variables to store the wind speed and direction, the mass of the projectile, and the drag
96 | Chapter 4: Kinetics
www.it-ebooks.info
coefficients. You’ll also have to add some controls in the dialog window so that you can
change these variables when you run the program. Figure 4-5 shows how we added
these interface controls in the upper-right corner of the main window.
Figure 4-5. Revised cannon example screenshot
We also added these lines to the DemoDlgProc function to handle the new wind speed
and direction values:
//-.//
LRESULT CALLBACK DemoDlgProc(HWND hDlg, UINT message, WPARAM wParam, LPARAM IParam)
//-.//
{
case WM INITDIALOG:
// New variables:
sprintf( str, "%f", m );
SetDlgItemText(hDlg, IDC_M, str);
sprintf( str, "%f", Cd );
SetDlgItemText(hDlg, IDC_CD, str);
Particle Kinetics in 3D | 97
www.it-ebooks.info
sprintf( str, "%f", Vw );
SetDlgItemText(hDlg, IDC_VW, str);
sprintf( str, "%f", GannaW );
SetDlgItemText(hDlg, IDC_GAMMAW, str);
sprintf( str, "%f", Cw );
SetDlgItenText(hDlg, IDC_CW, str);
case IDC_REFRESH:
// New variables:
GetDlgItenText(hDlg, IDC_M, str, 15);
m = atof(str);
GetDlgItenText(hDlg, IDC_CD, str, 15);
Cd = atof(str);
GetDlgItemText(hDlg, IDC_VW, str, 15);
Vw = atof(str);
GetDlgItemText(hDlg, IDC_GAMMAW, str, 15);
GammaW = atof(str);
GetDlgItenText(hDlg, IDC_CW, str, 15);
Cw = atof(str);
case IDC FIRE:
// New variables:
GetDlgItemText(hDlg, IDC_M, str, 15);
m = atof(str);
GetDlgItemText(hDlg, IDC_CD, str, 15);
Cd = atof(str);
GetDlgItemText(hDlg, IDC_VW, str, 15);
Vw = atof(str);
GetDlgItenText(hDlg, IDC_GAMMAW, str, 15);
GammaW = atof(str);
98 | Chapter 4: Kinetics
www.it-ebooks.info
GetDlgItemText(hDlg, IDC_Ckl, str, 15);
Cw = atof(str);
}
After playing with this example program, you should readily see that the trajectory of
the projectile is noticeably different from that typically obtained in the original example.
By adjusting the values of the wind speed, direction, and drag coefficients, you can
dramatically affect the projectile’s trajectory. If you set the wind speed to 0 and the drag
coefficients to 1, the trajectory will look like that obtained in the original example, where
wind and drag were not taken into account. Be careful, though: don’t set the drag co¬
efficient to 0 because this will result in a “divide by zero” error. We didn’t put the ex¬
ception handler in the program, but you can see that it will happen by looking at the
displacement vector formulas where the drag coefficient appears in the denominator of
several terms.
From a user’s perspective, if this were a video game, the problem of hitting the target
becomes much more challenging when wind and drag are taken into account. The wind
element is particularly interesting because you can change the wind speed and direction
during game play, forcing the user to pay careful attention to the wind in order to
accurately hit the target.
Rigid-Body Kinetics
You already know from your study of kinematics in Chapter 2 that dealing with rigid
bodies adds rotation, or angular motion, into the mix of things to consider. As we stated
earlier, the equations of motion now consist of a set of equations that relate forces to
linear accelerations and another set of equations that relate moments to angular accel¬
erations. Alternatively, you can think of the equations of motion as relating forces to
the rate of change in linear momentum, and moments to the rate of change in angular
momentum, as discussed in Chapter 1.
As in kinematics, the procedure for dealing with rigid-body kinetics problems involves
two distinct aspects: 1) tracking the translation of the body’s center of mass, where the
body is treated as a particle, and 2) tracking the body’s rotation, where you’ll utilize the
principles of local coordinates and relative angular velocity and acceleration, as dis¬
cussed in Chapter 2. Really, the only difference between rigid-body kinematics and
kinetics problems is that in kinetics problems we have forces to consider (including
their resulting moments).
The vector equations are repeated here for convenience:
F = ma
M cg = I a
Rigid-Body Kinetics | 99
www.it-ebooks.info
where in two dimensions:
XF = XF x i + XFy j
£F = <J(ZF x f + {ZF,f
Going from two-dimensional particle problems to two-dimensional rigid-body prob¬
lems involves only the addition of one more equation. This equation is, of course, the
moment equation relating the sum of all moments acting on the body to the body’s
moment of inertia and its angular acceleration. In plane motion, the axis of rotation of
the rigid body is always perpendicular to the coordinate plane. And since there is only
one axis of rotation, there is only one inertia term and one angular acceleration term to
consider. Thus, you can write:
M ca = I a
where M cg is the total moment and is calculated with the formulas discussed in the
section “Force and Torque” on page 80 in Chapter 3, and I is calculated about the axis
of rotation using the techniques discussed in the section “Mass, Center of Mass, and
Moment of Inertia” on page 9 in Chapter 1.
In their component forms, the set of equations of motion for two-dimensional kinetics
problems are:
XF x = m a x
XF y = m a y
XM cg = I a
Since these equations indicate linear motion on the xy-plane, the angular acceleration
will be about the z-axis perpendicular to the xy-plane. Likewise, the moment of inertia,
I, will be taken about the z-axis.
Recall from Chapter 3 that we calculate moment by taking the cross product of the
position vector for the force under consideration and the force vector. This means that,
unlike with particle kinetics, you now have to keep track of exactly where on the body
each force is applied. This is best illustrated with an example.
Consider the box of uniform density shown in Figure 4-6. Uniform density means that
its center of gravity is at the box’s geometric center. Find the value of the minimum force,
F , applied at the upper edge of the box, required to start tipping the box over.
100 | Chapter 4: Kinetics
www.it-ebooks.info
In Figure 4-6, F„ is the applied force, R, and R 2 are the reaction forces at supports one
and two, F fl and h' f2 are the forces due to friction at points one and two, and mg is the
weight of the box.
This is an example of the type of problem where you know something about the motion
of the object and have to find the value of one or more forces acting on it. To find the
value of the force that will be just enough to start tipping the box, you need to look at
the instant when the reaction force at support two is 0. This implies that all of the weight
of the box is now supported at point one and the box is starting to rotate over. At this
instant, just before it starts to rotate, the angular acceleration of the box is 0. Note that
the box’s linear acceleration isn’t necessarily 0—that is, you can push on the box and it
may slide without actually tipping over.
The equations of motion for this problem are:
£F X = F p - F f i - Fq = m a x
ZF y = Ri+R 2 -mg = ma y = 0
ZM cg = F p (h/2) + R 2 (w/2) - Ri (w/2) + F f2 (h/2) + F fl (h/2) = 1
a = 0
Rewriting the second preceding equation when R 2 is 0 shows that R 1 is equal to the
weight of the box. Further, when R 2 is 0, the R 2 (w/2) term drops out of the moment
equation, which can be rewritten by solving for F in terms of R v Note that when R,
goes to 0, so does F f2 . After some algebra, the equation looks as follows:
Rigid-Body Kinetics | 101
www.it-ebooks.info
F p = mg (w/h) - Ffi
Here you can see that the tipping force, applied to the upper edge, is proportional to the
weight and size of the box (actually the ratio of its width to its height), which you can
readily appreciate from a physical point of view. The friction term is important here
because the existence of the friction force actually helps the box to tip. If the box were
on a frictionless surface, it would tend to slide rather than tip.
In the case of plane motion, or 2D motion, of rigid bodies as discussed here, you are
able to readily set up the equations of motion and investigate both the linear as well as
angular motion of the body. In generalized three-dimensional motion, the linear motion
of rigid bodies is no different from that of particles; you simply track the motion of the
rigid body’s center of gravity. In three dimensions, however, rotation gives us some grief,
as it is no longer a simple matter of treating rotation about a single axis as in plane
motion. In 3D you’ll have to consider rotation about any axis, which leads to some
difficulties in representing arbitrary rotations (Euler angles won’t work for us) as well
as to complications with determining moments of inertia for rotation about any axis.
We’ll discuss these issues in Chapter 11 through Chapter 13.
102 | Chapter 4: Kinetics
www.it-ebooks.info
CHAPTER 5
Collisions
Now that you understand the motion of particles and rigid bodies, you need to consider
what happens when they run into each other. That’s what we’ll address in this chapter;
specifically, we’ll show you how to handle particle and, more interestingly, rigid-body
collision response.
Before moving forward, we need to make a distinction between collision detection and
collision response. Collision detection is a computational geometry problem involving
the determination of whether and where two or more objects have collided. Collision
response is a kinetics problem involving the motion of two or more objects after they
have collided. While the two problems are intimately related, we’ll focus solely on the
problem of collision response in this chapter. Later, in Chapter 7 through Chapter 13,
we’ll show you how to implement collision detection and response in various real-time
simulations, which draw upon concepts presented in this chapter.
Our treatment of rigid-body collision response in this chapter is based on classical
(Newtonian) impact principles. Here, colliding bodies are treated as rigid irrespective
of their construction and material. As in earlier chapters, the rigid bodies discussed here
do not change shape even upon impact. This, of course, is an idealization. You know
from your everyday experience that when objects collide they dent, bend, compress, or
crumple. For example, when a baseball strikes a bat, it may compress as much as three-
quarters of an inch during the millisecond of impact. Notwithstanding this reality, we’ll
rely on well-established analytical and empirical methods to approximate rigid-body
collisions.
This classical approach is widely used in engineering machine design, analysis, and
simulations; however, for rigid-body simulations there is another class of methods,
103
www.it-ebooks.info
known as penalty methods, at your disposal. 1 In penalty methods, the force at impact is
represented by a temporary spring that gets compressed between the objects at the point
of impact. This spring compresses over a very short time and applies equal and opposite
forces to the colliding bodies to simulate collision response. Proponents of this method
say it has the advantage of ease of implementation. However, one of the difficulties
encountered in its implementation is numerical instability. There are other arguments
for and against the use of penalty methods, but we won’t get into the debate here. Instead,
we’ve included several references in the Bibliography for you to review if you are so
inclined. Other methods of modeling collisions exist as well. For example, nonlinear
finite element simulations are commonly used to model collisions during product de¬
sign, such as the impact of a cellphone with the ground. These methods can be quite
accurate; however, they are too slow for real-time applications. Further, they are overkill
for games.
Impulse-Momentum Principle
Impulse is defined as a force that acts over a very short period of time. For example, the
force exerted on a bullet when fired from a gun is an impulse force. The collision forces
between two colliding objects are impulse forces, as when you kick a football or hit a
baseball with a bat.
More specifically, impulse is a vector quantity equal to the change in momentum. The
so-called impulse-momentum principle says that the change in moment is equal to the
applied impulse. For problems involving constant mass and moment of inertia, you can
write:
Linear impulse = 6 ( t _ t01+ ) F dt = m (v. - v_)
Angular impulse = 6 ( t _ t0 1 +) M dt = I (to+ - ©_)
In these equations, F is the impulsive force, M is the impulsive torque (or moment), t
is time, v is velocity, the subscript - refers to the instant just prior to impact, and the
subscript + refers to the instant just after impact. You can calculate the average impulse
force and torque using the following equations:
F = m (v 4 - v_) / (t + - t_)
M = I (® 4 - ©_) / (t+ - t_)
1. We use the classical approach in this book and are mentioning penalty methods only to let you know that the
method were going to show is not the only one. Roughly speaking, the penalty in penalty methods refers to
the numerical spring constants, which are usually large, that are used to represent the stiffness of the springs
and thus the hardness (or softness) of the colliding bodies. These constants are used in the system of equations
of motion describing the motion of all the bodies under consideration before and after the collision.
104 | Chapter 5: Collisions
www.it-ebooks.info
Consider this simple example: a 150 gram (0.15 kg) bullet is fired from a gun at a muzzle
velocity of756 m/s. The bullet takes 0.0008 seconds to travel through the 610 mm (0.610
m) rifle barrel. Calculate the impulse and the average impulsive force exerted on the
bullet. In this example, the bullets mass is a constant 150 grams and its initial velocity
is 0, thus its initial momentum is 0. Immediately after the gun is fired, the bullet’s mo¬
mentum is its mass times the muzzle velocity, which yields a momentum of 113.4 kg-
m/s. The impulse is equal to the change in momentum, and is simply 113.4 kg-m/s. The
average impulse force is equal to the impulse divided by the duration of application of
the force, or in this case:
Average impulse force = (113.4 kg-m/s) / (0.0008 s)
Average impulse force = 141,750 N
This is a simple but important illustration of the concept of impulse, and you’ll use the
same principle when dealing with rigid-body impacts. During impacts, the forces of
impact are usually very high and the duration of impact is usually very short. When two
objects collide, each applies an impulse force to the other; these forces are equal in
magnitude but opposite in direction. In the gun example, the impulse applied to the
bullet to set it in motion is also applied in the opposite direction to the gun, giving you
a nice kick in the shoulder. This is simply Newton’s third law in action.
Impact
In addition to the impulse momentum principle discussed in the previous section, our
classical impact, or collision response, analysis relies on another fundamental principle:
Newton’s principle of conservation of momentum, which states that when a system of
rigid bodies collide, momentum is conserved. This means that for bodies of constant
mass, the sum of their masses times their respective velocities before the impact is equal
to the sum of their masses times their respective velocities after the impact:
mivi_ + m 2 v 2 _ = mjvi + + m 2 v 2+
Here, m refers to mass, v refers to velocity, subscript 1 refers to body one, subscript 2
refers to body two, subscript - refers to the instant just prior to impact, and subscript
+ refers to the instant just after impact.
A crucial assumption of this method is that during the instant of impact the only force
that matters is the impact force; all other forces are assumed negligible over that very
short duration. Remember this assumption, because in Chapter 10 we’ll rely on it when
implementing collision response in an example 2D real-time simulation.
We’ve already stated that rigid bodies don’t change shape during impacts, and you know
from your own experience that real objects do change shape during impacts. What’s
happening in real life is that kinetic energy is being converted to strain energy to cause
Impact | 105
www.it-ebooks.info
the objects to deform. (See the sidebar “Kinetic Energy” on page 106 for further details on
this topic.) When the deformation in the objects is permanent, energy is lost and thus
kinetic energy is not conserved.
Kinetic Energy
Kinetic energy is a form of energy associated with moving bodies. It is equal to the energy
required to accelerate the body from rest, which is also equal to the energy required to
bring the moving body to a stop. As you might expect, kinetic energy is a function of
the body’s speed, or velocity, in addition to its mass. The formula for linear kinetic energy
is:
KE linear = (1/2) m v 2
Angular, or rotational, kinetic energy is a function of the body’s inertia and angular
velocity:
KE an g U i ar = (1/2) I (0“
Conservation of kinetic energy between two colliding bodies means that the sum of
kinetic energy of both bodies prior to impact is equal to the sum of the kinetic energy
of both bodies after impact:
miv 2 i_ + m 2 v 2 2 _ = m!V 2 1+ + m 2 v 2 2+
Collisions that involve losses in kinetic energy are said to be inelastic, or plastic, colli¬
sions. For example, if you throw two clay balls against each other, their kinetic energy
is converted to permanent strain energy in deforming the clay balls, and their collision
response—that is, their motion after impact—is less than spectacular. If the collision is
perfectly inelastic, then the two balls of clay will stick to each other and move together
at the same velocity after impact. Collisions where kinetic energy is conserved are called
perfectly elastic. In these collisions, the sum of kinetic energy of all objects before the
impact is equal to the sum of kinetic energy of all objects after the impact. A good
example of elastic impact (though not perfectly elastic) is the collision between two
billiard balls where the ball deformation is negligible and certainly not permanent under
normal circumstances.
Of course, in reality, impacts are somewhere between perfectly elastic and perfectly
inelastic. This means that for rigid bodies, which don’t change shape at all, we’ll have to
rely on an empirical relation to quantify the degree of elasticity of the impact(s) that
we’re trying to simulate. The relation that we’ll use is the ratio of the relative separation
velocity to the relative approach velocity of the colliding objects:
106 | Chapter 5: Collisions
www.it-ebooks.info
e = -(vi+ - v 2+ ) / (vi_ - v 2 _)
Here e is known as the coefficient of restitution and is a function of the colliding objects’
material, construction, and geometry. This coefficient can be experimentally deter¬
mined for specific impact scenarios—for example, the collision between a baseball and
bat, or a golf club and ball. For perfectly inelastic collisions, e is 0, and for perfectly elastic
collisions, e is 1. For collisions that are neither perfectly inelastic nor perfectly elastic, e
can be any value between 0 and 1. In this regard, the velocities considered are along the
line of action of the collision.
In frictionless collisions, the line of action of impact is a line perpendicular (or normal)
to the colliding surfaces. When the velocity of the bodies is along the line of action, the
impact is said to be direct. When the line of action passes through the center of mass of
the bodies, the collision is said to be central. Particles and spheres of uniform mass
distribution always experience central impact. Direct central impact occurs when the
line of action passes through the centers of mass of the colliding bodies and their velocity
is along the line of action. When the velocities of the bodies are not along the line of
action, the impact is said to be oblique. You can analyze oblique impacts in terms of
component coordinates where the component parallel to the line of action experiences
the impact, but the component perpendicular to the line of action does not. Figure 5-1
illustrates these impacts.
As an example, consider the collision between two billiard balls illustrated in Figure 5-2.
Impact | 107
www.it-ebooks.info
Figure 5-2. Example billiard ball collision
Both balls are a standard 57 mm in diameter, and each weighs 156 grams. Assume that
the collision is nearly perfectly elastic and the coefficient of restitution is 0.9. If the
velocity of ball 1 when it strikes ball 2 is 6 m/s in the x-direction, as shown in
Figure 5-2, calculate the velocities of both balls after the collision assuming that this is
a frictionless collision.
The first thing you need to do is recognize that the line of action of impact is along the
line connecting the centers of gravity of both balls, which, since these are spheres, is also
normal to both surfaces. You can then write the unit normal vector as follows:
V((2r) 2 -r 2 )i-rj
n = -2r-
n = (0.866) i - (0.5)j
where n is the unit normal vector, r is the ball radius, and i and j represent unit vectors
in the x- and y-directions, respectively.
Now that you have the line of action of the collision, or the unit normal vector, you can
calculate the relative normal velocity between the balls at the instant of collision.
v rn = [v i_ — v 2 _] • n
v rn = [(6 m/s) i + (0 m/s) j ] . [ (0.864) i - (0.5) j]
v rn = 5.18 m/s
This will be used as v ln _ in the following equations. Notice here that since ball 2 is initially
at rest, v 2 _ is 0.
Now you can apply the principle of conservation of momentum in the normal direction
as follows:
108 | Chapter 5: Collisions
www.it-ebooks.info
m l Vin- + m 2 V 2n - = m v ln+ + m 2 V 2n+
Noting that m 1 equals m 2 since the balls are identical, and that v 2n _ is 0, and then solving
for v ln+ yields:
v ln+ = v ln-“ v 2n+
To actually solve for these velocities, you need to use the equation for coefficient of
restitution and make the substitution for v ln+ . Then, you’ll be able to solve for v 2n+ . Here’s
how to proceed:
e = (—vin+ + v 2n+ ) / (v ln - v 2n _)
e v !n _ = —(v ln _ - v 2n+ ) + v 2n+
V2n+ = Vln- (e + 1) / 2
v 2n+ = (5.18 m/s)(1.9) / 2 = 4.92 m/s
Using this result and the formula for v ln+ yields:
vi n+ = 5.18 m/s -4.92 m/s = 0.26 m/s
Since the collision is frictionless, there is no impulse acting in the tangential direction.
This means that momentum is conserved in that direction too and that the final tan¬
gential speed of ball 1 is equal to its initial tangential speed, which in this case is equal
to 3 m/s (this equals (6m/s) sin 30°). Since ball 2 had no initial tangential speed, its
velocity after impact is solely in the normal direction. Converting these results back to
x-y coordinates instead of normal and tangential coordinates yields the following ve¬
locities for each ball after impact:
v 2+ = (4.92 m/s) sin 60° i - (4.92 m/s) cos 60 ° j
vi + = [(0.26 m/s) cos 30° + (3 m/s) sin 30°] i +
+ [(-0.26 m/s) sin 30° + (3 m/s) cos 30°)] j
vi + = (1.72 m/s) i + (2.47 m/s) j
To further illustrate the application of these collision response principles, consider an¬
other example, this time the collision between a baseball bat and baseball (as shown in
Figure 5-3). We are looking at a side view, staring down the barrel of the bat.
Impact | 109
www.it-ebooks.info
Figure 5-3. Example baseball and bat collision
To a reasonable degree of accuracy, the motion of a baseball bat at the instant of collision
can generally be described as independent of the batter—in other words, you can assume
that the bat is moving freely and pivoting about a point located near the handle end of
the bat. Assume that the ball strikes the bat on the sweet spot—that is, a point near
the center of percussion. 2 Further assume that the bat is swung in the horizontal plane
and that the baseball is traveling in the horizontal plane when it strikes the bat. The bat
is of standard dimensions with a maximum diameter of 70 mm and a weight of 1.02 kg.
The ball is also of standard dimensions with a radius of 37 mm and a weight of 0.15 kg.
The ball reaches a speed of 40 m/s (90 mph) at the instant it strikes the bat, and the speed
of the bat at the point of impact is 31 m/s (70 mph). For this collision, the coefficient of
restitution is 0.46. In the millisecond of impact that occurs, the baseball compresses
quite a bit; however, in this analysis assume that both the bat and the ball are rigid.
Finally, assume that this impact is frictionless.
As in the previous example, the line of action of impact is along the line connecting the
centers of gravity of the bat and ball; thus, the unit normal vector is:
_ -VVi + r 2 f - /y)i - rj
H ^
n = (0.875) i + (0.484) j
Here the subscripts 1 and 2 denote the bat and ball, respectively.
The relative normal velocity between the bat and ball is:
2. The center of percussion is a point located near one of the nodes of natural vibration, and is the point at
which, when the bat strikes the ball, no force is transmitted to the handle of the bat. If you’ve ever hit a baseball
incorrectly such that you get a painful vibrating sensation in your hands, then you know what it feels like to
miss the center of percussion.
110 | Chapter 5: Collisions
www.it-ebooks.info
v rn = [v 1 — - v 2 _] • n
V rn = [(71 m/s) i + (0 m/s) j ] • [ (0.875) i - (0.484) j]
v rn = 62.1 m/s
The velocity components of the bat and ball in the normal direction are:
vj n — = vi_ • n = 27.1 m/s
v 2n _ = v 2 _ • n = -35.0 m/s
Applying the principle of conservation of momentum in the normal direction and
solving for v ln+ yields:
mi v !n _ + m 2 v 2n _ = m! v !n+ + m 2 v 2n+
(1.02 kg) (27.1 m/s) + (0.15 kg) (-35.2 m/s) =
=(1.02 kg) vi n _|_ + (0.15 kg) v 2n+
vi n+ = 21.92 m/s - (0.14 m/s) v 2n+
As in the previous example, applying the formula for coefficient of restitution with the
preceding formula for v ln+ yields:
e = (—vi n+ + v 2n+ ) / (v ln _ - v 2n _)
0.46 = [-21.92 m/s + (0.14 m/s) v 2n+ + v 2n+ ] / [27.1 m/s + 35.2
m/s]
v 2n+ = 44.4 m/s and vi n+ = 15.7 m/s
Here again, since this impact is frictionless, each object retains its original tangential
velocity component. For the bat, this component is 15 m/s, while for the ball it’s -19.3
m/s. Converting these normal and tangential components to x-y coordinates yields the
following bat and ball velocities for the instant just after impact:
V] + = 21.0 m/s i - 5.5 m/s j
v 2+ = 30 m/s i + 38.7 m/s j
Both of these examples illustrate fundamental impact analysis using the classical ap¬
proach. They also share an important assumption—that the impacts are frictionless. In
reality, you know that billiard balls and baseballs and bats collide with friction; other¬
wise, you would not be able to apply English in billiards or create lift-generating spin
on baseballs. Later in this chapter we’ll discuss how to include friction in your impact
analysis.
Impact | 111
www.it-ebooks.info
Linear and Angular Impulse
In the previous section, you were able to work through the specific examples by hand
using the principle of conservation of momentum and the coefficient of restitution. This
approach will suffice if you’re writing games where the collision events are well defined
and anticipated. However, if you’re writing a real-time simulation where objects, espe¬
cially arbitrarily shaped rigid bodies, may or may not collide, then you’ll want to use a
more general approach. This approach involves the use of formulas to calculate the
actual impulse between colliding objects so that you can apply this impulse to each
object, instantly changing its velocity. In this section, we’ll derive the equations for
impulse, both linear and angular, and we’ll show you how to implement these equations
in code in Chapter 10.
When you’re dealing with particles or spheres, the only impulse formula that you’ll need
is that for linear impulse, which will allow you to calculate the new linear velocities of
the objects after impact. So, the first formula that we’ll derive for you is that for linear
impulse between two colliding objects, as shown in Figure 5-4.
velocity after impact
Body 1
velocity before impact
Figure 5-4. Two colliding particles (or spheres)
For now, assume the collision is frictionless and the line of action of the impulse is along
the line connecting the centers of mass of the two objects. This line is normal to the
surfaces of both objects.
To derive the formula for linear impulse, you have to consider the formula from the
definition of impulse along with the formula for coefficient of restitution. Here let J
represent the impulse:
IJI = m (lv+1 - Iv J)
112 | Chapter 5: Collisions
www.it-ebooks.info
e = —(lv 1+ l - lv 2 +l)/(lvi_l - lv 2 J)
In these equations the velocities are those along the line of action of the impact, which
in this case is a line connecting the centers of mass of the two objects. Since the same
impulse applies to each object (just in opposite directions), you actually have three
equations to deal with:
IJI = mj (lvi+1 - lvi_l)
l-JI = m 2 (lv 2+ l - Iv 2 —I)
e = —(lv 1+ l— lv 2 +l)/(lvi_l - lv 2 J)
Notice we’ve assumed that / acts positively on body 1 and its negation, -/, acts on body
2. Also notice that there are three unknowns in these equations: the impulse and the
velocities of both bodies after the impact. Since there are three equations and three
unknowns, you can solve for each unknown by rearranging the two impulse equations
and substituting them into the equation for e. After some algebra, you’ll end up with a
formula for / that you can then use to determine the velocities of each body just after
impact. Here’s how that’s done:
For body 1: lvi+1 = IJI/mj + lvi_l
For body 2: lv 2 +l = —IJI/m 2 + lv 2 J
Substituting |v 1+ | and |v 2+ | into the equation for e yields:
e (IviJ - Iv 2 _I) = -[(IJI/mi + IviJ) - (-IJI/m 2 + lv 2 _l)]
e (Ivi-I - lv 2 _l) + lvi_l - lv 2 _I = -J (1/mj + l/m 2 )
Let |v r | = (|Vj_| - |v 2 _|); then:
e lv r l + lv r l = -IJI (1/nq + l/m 2 )
IJI =-lv r l(e+ l)/(l/mj + l/m 2 )
Since the line of action is normal to the colliding surfaces, v r is the relative velocity along
the line of action of impact, and J acts along the line of action of impact, which in this
case is normal to the surfaces, as we’ve already stated.
Now that you have a formula for the impulse, you can use the definition of impulse
along with this formula to calculate the change in linear velocity of the objects involved
in the impact. Here’s how that’s done in the case of two objects colliding:
vi+ = vi_ + (IJI n)/mi
v 2+ = v 2 _ + (-IJI n)/m 2
Linear and Angular Impulse | 113
www.it-ebooks.info
Notice that for the second object, the negative of the impulse is applied since it acts on
both objects equally but in opposite directions.
When dealing with rigid bodies that rotate, you’ll have to derive a new equation for
impulse that includes angular effects. You’ll use this impulse to calculate new linear and
angular velocities of the objects just after impact. Consider the two objects colliding at
point P, as shown in Figure 5-5.
There’s a crucial distinction between this collision and that discussed earlier. In this case,
the velocity at the point of contact on each body is a function of not only the objects’
linear velocity but also their angular velocities, and you’ll have to recall from Chap¬
ter 2 the following formula in order to calculate the velocities at the impact point on
each body:
v p = v g + (w X r)
In this relation, r is the vector from the body’s center of gravity to the point P.
Using this formula, you can rewrite the two formulas relating the linear velocity after
impact to the impulse and initial velocity as follows:
For body 1: vi g+ + (a>i+ X iq) = J/mi + vi g _ + (®i_ X iq)
For body 2: v 2g + + (w 2+ X r 2 ) = -J/m 2 + v 2g _ + (<o 2 _ x r 2 )
There are two additional unknowns here, the angular velocities after impact, which
means that you need two additional equations. You can get these equations from the
definition of angular impulse:
For body 1: (iq X J) = Ii (®i + - ®i_)
For body 2: (r 2 X -J) = I 2 (to 2+ - w 2 _)
114 | Chapter 5: Collisions
www.it-ebooks.info
Here we calculate the moment due to the impulse by taking the vector cross product of
the impulse with the distance from the body’s center of gravity to the point of application
of the impulse.
By combining all of these equations with the equation for e and following the same
procedure used when deriving the linear impulse formula, you’ll end up with a formula
for | J| that takes into account both linear and angular effects, which you can then use
to find the linear and angular velocities of each body immediately after impact. Here’s
the result:
IJI = -(v r « n)(e + l)/[l/mj + l/m 2 + n «((ri X n)/Ij) x rj + n •
((r 2 X n)/I 2 ) X r 2 ]
Here v r is the relative velocity along the line of action at the impact point P, and n is a
unit vector along the line of action at the impact point pointing out from body 1.
With this new formula for |J|, you can calculate the change in linear and angular ve¬
locities of the objects involved in the collision using these formulas:
vi+ = Vi_ + (IJI n)/mi
v 2+ = v 2 _ + (-IJI n)/m 2
© 1 + = g>i_ + (ri X IJI n)/Ii
©2+ = <*>2— + ( r 2 X -IJI n)/I 2
As we said earlier, we’ll show you how to implement these formulas for impulse in code
when you get to Chapter 10.
Friction
Friction acts between contacting surfaces to resist motion. When objects collide in any
type of collision except direct impact, for that very brief moment of contact, they will
experience a friction force that acts tangentially to the contacting surfaces. Not only will
this tangential force change the linear velocities of the colliding objects in the tangential
direction, but it will also create a moment (torque) on the obj ects, which tends to change
their angular velocities. This tangential impulse combined with the normal impulse
results in an effective line of action of the total collision impulse that is no longer per¬
pendicular to the contacting surfaces.
In practice, it is very difficult to quantify this collision friction force due to the fact that
the friction force is not necessarily constant if the collision is such that the friction force
does not develop beyond the maximum static friction force. Further complications stem
from the fact that objects do tend to deform when they collide, creating an additional
source of resistance. That said, since the friction force is a function of the normal force
between the contacting surfaces, you know that the ratio of the normal force to the
Friction | 115
www.it-ebooks.info
friction force is equal to the coefficient of friction. If you assume that the collisions are
such that the kinetic coefficient of friction is applicable, then this ratio is constant.
hk = Ff/Fn
Here, f' t is the tangential friction force and F n is the normal impact force. You can extend
this to say that the ratio of the tangential impulse to normal impulse is equal to the
coefficient of friction.
Consider the collision between the golf club head and golf ball illustrated in Figure 5-6.
Figure 5-6. Golf club-golf ball collision
In the upper velocity diagram, v_ represents the relative velocity between the ball and
club head at the instant of impact, v + represents the velocity of the ball just after impact,
v t _ and v t+ represent the tangential components of the ball velocity at and just after the
instant of impact, respectively.
If this were a frictionless collision, v t _ and v t+ would be equal, as would the angles a and
0. However, with friction the tangential velocity of the ball is reduced, making v t+ less
than v t _, which also means that a will be less than 0.
The lower force diagram in Figure 5-6 illustrates the forces involved in this collision
with friction. Since the ratio of the tangential friction force to the normal collision force
is equal to the coefficient of friction, you can develop an equation relating the angle cp
to the coefficient of friction.
tan cp = Ff / F n = p
In addition to this friction force changing the linear velocity of the ball in the tangential
direction, it will also change the angular velocity of the ball. Since the friction force is
116 | Chapter 5: Collisions
www.it-ebooks.info
acting on the ball’s surface some distance from its center of gravity, it creates a moment
(torque) about the ball’s center of gravity, which causes the ball to spin. You can develop
an equation for the new angular velocity of the ball in terms of the normal impact force
or impulse:
£ M cg = Ff r = I C g dco/dt
p F n r = I cg dco/dt
p F n r dt = I cg dco
6(t- to t+) Fn dt = I C g / (p r) to (o+) ® d (0
Notice here that the integral on the left is the normal impulse; thus:
Impulse = I eg / (p r) (co + - co_)
co + = (Impulse) (p r) / I cg + co_
This relation looks very similar to the angular impulse equation that we showed you
earlier in this chapter, and you can use it to approximate the friction-induced spin in
specific collision problems.
Turn back to the equation for impulse, /, in the last section that includes both linear and
angular effects. Here it is again for convenience:
IJI = -(v r * n)(e + 1) / [1/mj + l/m 2 + n *((rj x n)/Ij) x iq + n •
((r 2 X n)/I 2 ) X r 2 ]
This formula gives you the collision impulse in the normal direction. To see how friction
fits in you must keep in mind that friction acts tangentially to the contacting surfaces,
that combining the friction force with the normal impact force yields a new effective
line of action for the collision, and that the friction force (and impulse) is a function of
the normal force (impulse) and coefficient of friction. Considering all these factors, the
new equations to calculate the change in linear and angular velocities of two colliding
objects are as follows:
Vi+ = vi_ + (J n + (p J) t) / mi
v 2+ = v 2 _ + (-J n + (p J) t) / m 2
w 1+ = «i- + (iq x (J n+ (p J) t)) / leg
®2+ = < 0 2 _ + (r 2 X (-J n+ (p J) t)) / ^
In these equations, t is the unit tangent vector, which is tangent to the collision surfaces
and at a right angle to the unit normal vector. You can calculate the tangent vector if
you know the unit normal vector and the relative velocity vector in the same plane as
the normal vector.
Friction | 117
www.it-ebooks.info
t = (n x v r ) x n
t = t / Itl
For many problems that you’ll face, you may be able to reasonably neglect friction in
your collision response routines since its effect may be small compared to the effect of
the normal impulse itself. However, for some types of problems, friction is crucial. For
example, the flight trajectory of a golf ball depends greatly on the spin imparted to it as
a result of the club-ball collision. We’ll discuss how spin affects trajectory in the next
chapter, which covers projectile motion.
118 | Chapter 5: Collisions
www.it-ebooks.info
CHAPTER 6
Projectiles
This chapter is the first in a series of chapters that discuss specific real-world phenomena
and systems, such as projectile motion and airplanes, with the goal of giving you a solid
understanding of their real-life behavior. This understanding will help you to model
these or similar systems accurately in your games. Instead of relying on purely idealized
formulas, we’ll present a wide variety of practical formulas and data that you can use.
We’ve chosen the examples in this and later chapters to illustrate common forces and
phenomena that exists in many systems, not just the ones we’ll be discussing here. For
example, while Chapter 16, “Ships and Boats,” discusses buoyancy in detail, buoyancy
is not limited to ships; any object immersed in a fluid experiences buoyant forces. The
same applies for the topics discussed in this chapter and Chapter 15, Chapter 17, Chap¬
ter 18, and Chapter 19.
Once you understand what’s supposed to happen with these and similar systems, you’ll
be in a better position to interpret your simulation results to determine if they make
sense—that is, if they are realistic enough. You’ll also be better educated on what factors
are most important for a given system such that you can make appropriate simplifying
assumptions to help ease your effort. Basically, when designing and optimizing your
code, you’ll know where to cut things out without sacrificing realism. This gets into the
subject of parameter tuning.
Over the next few chapters, we want to give you enough of an understanding of certain
physical phenomenon such that you can tune your models for the desired behavior. If
you are modeling several similar objects in your simulation but want each one to behave
slightly differently, then you have to tune the forces that get applied to each object in
order to achieve the varying behavior. Since forces govern the behavior of objects in
your simulations, we’ll be focusing on force calculations with the intent of showing you
how and why certain forces are what they are instead of simply using the idealized
formulas discussed in Chapter 3. Parameter tuning isn’t just limited to tuning your
model’s behavior—it also involves dealing with numerical issues, such as numerical
119
www.it-ebooks.info
stability in your integration algorithms. We’ll discuss these issues more when we show
you several simulation examples in Chapter 7 through Chapter 14.
We’ve devoted this entire chapter to projectile motion because so many physical prob¬
lems that may find their way into your games fall in this category. Further, the forces
governing projectile motion affect many other systems that aren’t necessarily projectiles
—for example, the drag force experienced by projectiles is similar to that experienced
by airplanes, cars, or any other object moving through a fluid such as air or water.
A projectile is an object that is placed in motion by a force acting over a very short period
of time, which, as you know from Chapter 5, is also called an impulse. After the proj ectile
is set in motion by the initial impulse during the launching phase, the projectile enters
into the projectile motion phase, where there is no longer a thrust or propulsive force
acting on it. As you know already from the examples presented in Chapter 2 and Chap¬
ter 4, there are other forces that act on projectiles. (For the moment, we’re not talking
about self-propelled “projectiles” such as rockets since, due to their propulsive force,
they don’t follow “classical” projectile motion until after they’ve expended their fuel.)
In the simplest case, neglecting aerodynamic effects, the only force acting on a proj ectile
other than the initial impulsive force is gravitation. For situations where the projectile
is near the earth’s surface, the problem reduces to a constant acceleration problem.
Assuming that the earth’s surface is flat—that is, that its curvature is large compared to
the range of the projectile—the following statements describe projectile motion:
• The trajectory is parabolic.
• The maximum range, for a given launch velocity, occurs when the launch angle is
45°.
• The velocity at impact is equal to the launch velocity when the launch point and
impact point are at the same level.
• The vertical component of velocity is 0 at the apex of the trajectory.
• The time required to reach the apex is equal to the time required to descend from
the apex to the point of impact assuming that the launch point and impact point
are at the same level.
• The time required to descend from the apex to the point of impact equals the time
required for an object to fall the same vertical distance when dropped straight down
from a height equal to the height of the apex.
Simple Trajectories
There are four simple classes of projectile motion problems that we’ll summarize:
• When the target and launch point are at the same level
120 | Chapter 6: Projectiles
www.it-ebooks.info
• When the target is at a level higher than the launch point
• When the target is at a level lower than the launch point
• When the projectile is dropped from a moving system (like an airplane) above the
target
In the first type of problem, the launch point and the target point are located on the
same horizontal plane. In Figure 6-1, v 0 is the initial velocity of the projectile at the time
of launch, cp is the launch angle, R is the range of the projectile, and h is the height of
the apex of the trajectory.
R
Figure 6-1. Target and launch point at same level
To solve this type of problem, use the formulas shown in Table 6-1. Note, in these
formulas t represents any time instant after launch, and T represents the total time from
launch to impact.
Table 6-1. Formulas—target and launch point at same level
1 To calculate:
Use this formula: 1
x(t)
(v 0 COS cp) t
y(t)
(v 0 sin <p) t-(g t 2 ) / 2
VxW
v 0 cos cp
Vy(t)
v 0 sin cp — g t
v(t)
Jv 0 2 - 2gtv 0 sin ip + g 2 t 2
h
(v 0 2 sin 2 cp) / (2 g)
R
v 0 T cos cp
T
(2 v 0 sin cp) / g
Remember to keep your units consistent when applying these formulas. If you are
working in the SI (metric) system, length and distance values should be in meters (m);
Simple Trajectories | 121
www.it-ebooks.info
time should be in seconds (s); speed should be in meters per second (m/s); and accel¬
eration should be in meters per second squared (m/s 2 ). In the SI system, g is 9.8 m/s 2 .
In the second type of problem, the launch point is located on a lower horizontal plane
than the target. In Figure 6-2, the launch points y coordinate is lower than the targets
y coordinate.
Figure 6-2. Target higher than launch point
For this type of problem, use the formulas shown in Table 6-2. Notice that most of these
formulas are the same as those shown in Table 6-1.
Table 6-2. Formulas—target higher than launch point
1 To calculate:
Use this formula: I
<t)
(v„ cos <p)t
y(t)
(v 0 sin <p) t-(g t 2 ) / 2
v x (t)
V 0 COS (p
Vy(t)
v 0 sin (p — g t
v(t)
^v 0 2 - 2 gtv 0 sin <p + g 2 t 2
h
(v 0 2 sin 2 cp) / (2 g)
R
v 0 T cos cp
T
(v 0 sin cp)/g + V ,
Actually, the only formula that has changed is that for T, where it has been revised to
account for the difference in elevation between the target and the launch point.
In the third type of problem, the target is located on a plane lower than the launch point;
in Figure 6-3, the target’s y coordinate is lower than the launch point’s y coordinate.
122 | Chapter 6: Projectiles
www.it-ebooks.info
R
Figure 6-3. Target lower than launch point
Table 6-3 shows the formulas to use for this type of problem. Here again, almost all of
the formulas are the same as those shown in Table 6-1.
Table 6-3. Formulas—target lower than launch point
1 To calculate:
Use this formula: 1
x(t)
(v„ cos <p)t
y(t)
(v 0 sin <p) t-(g t 2 ) / 2
v x (t)
v„ cos cp
Vy(t)
v 0 sin <p — g t
v(t)
-y/r 0 2 - 2 gtv 0 sin tp + gh 2
h
b + (v 0 2 sin 2 cp) / (2 g)
R
v 0 T cos cp
T
(v 0 sin cp) / g + '/y-
The only formulas that have changed are the formulas for h and T, which have been
revised to account for the difference in elevation between the target and the launch point
(except this time the target is lower than the launch point).
Finally, the fourth type of problem involves dropping the projectile from a moving
system, such as an airplane. In this case, the initial velocity of the projectile is horizontal
and equal to the speed of the vehicle dropping it. Figure 6-4 illustrates this type of
problem.
Simple Trajectories | 123
www.it-ebooks.info
Figure 6-4. Projectile dropped from a moving system
Table 6-4 shows the formulas to use to solve this type of problem. Note here that when
v D is 0, the problem reduces to a simple free-fall problem in which the projectile drops
straight down.
Table 6-4. Formulas—projectile dropped from a moving system
1 To calculate:
Use this formula: 1
x(t)
v 0 t
y(t)
h — (g t 2 ) / 2
Vx(t)
Vo
Vy(t)
-gt
v(t)
fvj + g 2 t 2
h
(g t 2 ) / 2
R
v„T
T
These formulas are useful if you’re writing a game that does not require a more accurate
treatment of projectile motion—that is, if you don’t need or want to consider the other
forces that can act on a projectile when in motion. If you are going for more accuracy,
you’ll have to consider these other forces and treat the problem as we did in Chap¬
ter 4’s example.
Drag
In Chapter 3 and Chapter 4, we showed you the idealized formulas for viscous fluid
dynamic drag as well as how to implement drag in the equations of motion for a pro-
124 | Chapter 6: Projectiles
www.it-ebooks.info
jectile. This was illustrated in the example program discussed in Chapter 4. Recall that
the drag force is a vector just like any other force and that it acts on the line of action of
the velocity vector but in a direction opposing velocity. While those formulas work in
a game simulation, as we said before, they don’t tell the whole story. While we can’t treat
the subject of fluid dynamics in its entirety in this book, we do want to give you a better
understanding of drag than just the simple idealized equation presented earlier.
Analytical methods can show that the drag on an object moving through a fluid is
proportional to its speed, size, shape, and the density and viscosity of the fluid through
which it is moving. You can also come to these conclusions by drawing on your own
real-life experience. For example, when waving your hand through the air, you feel very
little resistance; however, if you put your hand out of a car window traveling at 100 km/
h, then you feel much greater resistance (drag force) on your hand. This is because drag
is speed dependent. When you wave your hand underwater—say, in a swimming pool
—you’ll feel a greater drag force on your hand than you do when waving it in the air.
This is because water is more dense and viscous than air. As you wave your hand un¬
derwater, you’ll notice a significant difference in drag depending on the orientation of
your hand. If your palm is in line with the direction of motion—that is, you are leading
with your palm—then you’ll feel a greater drag force than you would if your hand were
turned 90 degrees as though you were executing a karate chop through the water. This
tells you that drag is a function of the shape of the object. You get the idea.
To facilitate our discussion of fluid dynamic drag, let’s look at the flow around a sphere
moving through a fluid such as air or water. If the sphere is moving slowly through the
fluid, the flow pattern around the sphere would look something like Figure 6-5.
Figure 6-5. Flow pattern around slowly moving sphere
Bernoulli’s equation, which relates pressure to velocity in fluid flow, says that as the fluid
moves around the sphere and speeds up, the pressure in the fluid (locally) will go down.
The equation, presented by Daniel Bernoulli in 1738, applies to frictionless incompres¬
sible fluid flow and looks like this: 1
P/y + z + V 2 / (2g) = constant
1. In a real fluid with friction, this equation will have extra terms that account for energy losses due to friction.
Drag | 125
www.it-ebooks.info
where P is the pressure at a point in the fluid volume under consideration, y is the specific
weight of the fluid, z is the elevation of the point under consideration, V is the fluid
velocity at that point, and g is the acceleration due to gravity. As you can see, if the
expression on the left is to remain constant, and assuming that z is constant, then if
velocity increases the pressure must decrease. Likewise, if pressure increases, then ve¬
locity must decrease.
As you can see in Figure 6-5, the pressure will be greatest at the stagnation point, Sj, and
will decrease around the leading side of the sphere and then start to increase again
around the back of the sphere. In an ideal fluid with no friction, the pressure is fully
recovered behind the sphere and there is a trailing stagnation point, S„ whose pressure
is equal to the pressure at the leading stagnation point. Since the pressure fore and aft
of the sphere is equal and opposite, there is no net drag force acting on the sphere.
The pressure on the top and bottom of the sphere will be lower than at the stagnation
points since the fluid velocity is greater over the top and bottom. Since this is a case of
symmetric flow around the sphere, there will be no net pressure difference between the
top and bottom of the sphere.
In a real fluid there is friction, which affects the flow around the sphere such that the
pressure is never fully recovered on the aft side of the sphere. As the fluid flows around
the sphere, a thin layer sticks to the surface of the sphere due to friction. In this boundary
layer , the speed of the fluid varies from 0 at the sphere surface to the ideal free stream
velocity, as illustrated in Figure 6-6.
Figure 6-6. Velocity gradient within boundary layer
This velocity gradient represents a momentum transfer from the sphere to the fluid and
gives rise to the frictional component of drag. Since a certain amount of fluid is stick¬
ing to the sphere, you can think of this as the energy required to accelerate the fluid and
move it along with the sphere. (If the flow within this boundary layer is laminar, then
the viscous shear stress between fluid “layers” gives rise to friction drag. When the flow
is turbulent, the velocity gradient and thus the transfer of momentum gives rise to
friction drag.)
126 | Chapter 6: Projectiles
www.it-ebooks.info
Moving further aft along the sphere, the boundary layer grows in thickness and will not
be able to maintain its adherence to the sphere surface, and it will separate at some point.
Beyond this separation point, the flow will be turbulent, and this is called the turbulent
wake. In this region, the fluid pressure is lower than that at the front of the sphere. This
pressure differential gives rise to the pressure component of drag. Figure 6-7 shows how
the flow might look.
Figure 6-7. Flow pattern around sphere showing separation
For a slowly moving sphere, the separation point will be approximately 80° from the
leading edge.
Now, if you were to roughen the surface of the sphere, you’ll affect the flow around it.
As you would expect, this roughened sphere will have a higher friction drag component.
However, more importantly, the flow will adhere to the sphere longer and the separation
point will be pushed further back to approximately 115°, as shown in Figure 6-8.
Figure 6-8. Flow around a roughened sphere
This will reduce the size of the turbulent wake and the pressure differential, thus de¬
creasing the pressure drag. It’s paradoxical but true that, all other things being equal, a
slightly roughened sphere will have less total drag than a smooth one. Ever wonder why
golf balls have dimples? If so, there’s your answer.
The total drag on the sphere depends very much on the nature of the flow around the
sphere—that is, whether the flow is laminar or turbulent. This is best illustrated by
looking at some experimental data. Figure 6-9 shows a typical curve of the total drag
coefficient for a sphere plotted as a function of the Reynolds number.
Drag | 127
www.it-ebooks.info
Figure 6-9. Total drag coefficient for a smooth sphere versus Reynolds number 2
The Reynolds number (commonly denoted N t or R n ) is a dimensionless number that
represents the speed of fluid flow around an object. It’s a little more than just a speed
measure, since it includes a characteristic length for the object and the viscosity and
density of the fluid. The formula for the Reynolds number is:
R n = (v L)/o
or,
R n = (v L p)/p
where v is speed, L is a characteristic length of the object (diameter for a sphere), v is
the kinematic viscosity of the fluid, p is the fluid mass density, and p is the absolute
viscosity of the fluid. For the Reynolds number to work out as a dimensionless number
velocity, length and kinematic viscosity must have units of m/s, m, and m 2 /s, respectively,
within the SI system.
This number is useful for non-dimensionalizing data measured from tests on an object
of given size (like a model) such that you can scale the data to estimate the data for
similar objects of different size. Here “similar” means that the objects are geometrically
similar, just different scales, and that the flow pattern around the objects is similar. For
a sphere the characteristic length is diameter, so you can use drag data obtained from a
small model sphere of given diameter to estimate the drag for a larger sphere of different
2. The curve shown here is intended to demonstrate the trend of Cj versus R n for a smooth sphere. For more
accurate drag coefficient data for spheres and other shapes, refer to any college-level fluid mechanics text, such
as Robert L. Daugherty, Joseph B. Franzini, and E. John Finnemore’s Fluid Mechanics with Engineering Ap¬
plications (McGraw-Hill).
128 | Chapter 6: Projectiles
www.it-ebooks.info
diameter. A more useful application of this scaling technique is estimating the viscous
drag on ship or airplane appendages based on model test data obtained from wind tunnel
or tow tank experiments.
The Reynolds number is used as an indicator of the nature of fluid flow. A low Reynolds
number generally indicates laminar flow, while a high Reynolds number generally in¬
dicates turbulent flow. Somewhere in between, there is a transition range where the flow
makes the transition from laminar to turbulent flow. For carefully controlled experi¬
ments, this transition ( critical ) Reynolds number can consistently be determined. How¬
ever, in general the ambient flow field around an object—that is, whether it has low or
high turbulence—will affect when this transition occurs. Further, the transition Rey¬
nolds number is specific to the type of problem being investigated (for example, whether
you’re looking at flow within pipes, the flow around a ship, or the flow around an air¬
plane, etc.).
We calculate the total drag coefficient, C d , by measuring the total resistance, R t , from
tests and using the following formula:
C d = R t / (0.5 p v 2 A)
where A is a characteristic area that depends on the object being studied. For a sphere,
A is typically the projected frontal area of the sphere, which is equal to the area of a
circle of diameter equal to that of the sphere. By comparison, for ship hulls, A is typically
taken as the underwater surface area of the hull. If you work out the units on the right-
hand side of this equation, you’ll see that the drag coefficient is nondimensional (i.e., it
has no units).
Given the total drag coefficient, you can estimate the total resistance (drag) using the
following formula:
R t = (0.5 p v 2 A) C d
This is a better equation to use than the ones given in Chapter 3, assuming you have
sufficient information available—namely, the total drag coefficient, density, velocity,
and area. Note the dependence of total resistance on velocity squared. To get R t in units
of newtons (N), you must have velocity in m/s, area in m 2 , and density in kg/m 3 (re¬
member C d is dimensionless).
Turning back now to Figure 6-9, you can make a couple of observations. First, you can
see that the total drag coefficient decreases as the Reynolds number increases. This is
due to the formation of the separation point and its subsequent move aft on the sphere
as the Reynolds number increases and the relative reduction in pressure drag, as dis¬
cussed previously. At a Reynolds number of approximately 250,000, there is a dramatic
Drag | 129
www.it-ebooks.info
reduction in drag. This is a result of the flow becoming fully turbulent with a corre¬
sponding reduction in pressure drag.
In the Cannon2 example in Chapter 4, we implemented the ideal formula for air drag on
the proj ectile. In that case we used a constant value of drag coefficient that was arbitrarily
defined. As we said earlier, it would be better to use the formula presented in this chapter
for total drag along with the total drag coefficient data shown in Figure 6-9 to estimate
the drag on the projectile. While this is more “accurate,” it does complicate matters for
you. Specifically, the drag coefficient is now a function of the Reynolds number, which
is a function of velocity. You’ll have to set up a table of drag coefficients versus the
Reynolds number and interpolate this table given the Reynolds number calculated at
each time step. As an alternative, you can fit the drag coefficient data to a curve to derive
a formula that you can use instead; however, the drag coefficient data may be such that
you’ll have to use a piecewise approach and derive curve fits for each segment of the
drag coefficient curve. The sphere data presented herein is one such case. The data does
not lend itself nicely to a single polynomial curve fit over the full range of the Reynolds
number. In such cases, you’ll end up with a handful of formulas for drag coefficient,
with each formula valid over a limited range of Reynolds numbers.
While the Cannon2 example does have its limitations, it is useful to see the effects of drag
on the trajectory of the projectile. The obvious effect is that the trajectory is no longer
parabolic. You can see in Figure 6-10 that the trajectory appears to drop off much more
sharply when the projectile is making its descent after reaching its apex height.
Another important effect of drag on trajectory (this applies to objects in free fall as well)
is the fact that drag will limit the maximum vertical velocity attainable. This limit is the
so-called terminal velocity. Consider an object in free fall for a moment. As the object
accelerates toward the earth at the gravitation acceleration, its velocity increases. As
130 | Chapter 6: Projectiles
www.it-ebooks.info
velocity increases, so does drag since drag is a function of velocity. At some speed the
drag force retarding the object’s motion will increase to a point where it is equal to the
gravitational force that’s pulling the object toward the earth. In the absence of any other
forces that may affect motion, the net acceleration on the object is 0, and it continues
its descent at the constant terminal velocity.
Let us illustrate this further. Go back to the formula we derived for the y component
(vertical component) of velocity for the projectile modeled in the Cannon2 example.
Here it is again so you don’t have to flip back to Chapter 4:
v y2 = (1 / C d ) e(- c d /m k (C d v y i + m g) - (m g) / C d
It isn’t obvious from looking at this equation, but the velocity component, v y2 , asymptotes
to some constant value as time increases. To help you visualize this, we’ve plotted this
equation, as shown in Figure 6-11.
As you can see, over time the velocity reaches a maximum absolute value of about
-107.25 speed units. The negative velocities indicate that the velocity is in the negative
y-direction—that is, the object is falling toward the earth in this case. (For this calcu¬
lation we arbitrarily assumed a mass of 100, a drag coefficient of 30, and an initial velocity
of 0.)
Assuming an initial velocity of 0 and equating the formula for total resistance shown
earlier to the weight of an object, you can derive the following formula for terminal
velocity for an object in free fall:
v t
I -mg
y C d pA
Drag | 131
www.it-ebooks.info
The trick in applying this formula is in determining the right value for the drag coeffi¬
cient. Just for fun, let’s assume a drag coefficient of 0.5 and calculate the terminal velocity
for several different objects. This exercise will allow you to see the influence of the
object’s size on terminal velocity. Table 6-5 gives the terminal velocities for various
objects in free fall using an air density of 1.225 kg/m 3 (air at standard atmospheric
pressure at 15°C). Using this equation with density in kg/m 3 means that m must be in
kg, g in m/s 2 , and A in m 2 in order to get the terminal speed in m/s. We went ahead and
converted from m/s to kilometers per hour (km/h) to present the results in Table 6-5.
The weight of each object shown in this table is simply its mass, m, times g.
Table 6-5. Terminal velocities for various objects
1 Object
Weight (N)
Area (m 2 )
Terminal velocity (km/h) 1
Skydiver in free fall
801
0.84
201
Skydiverwith open parachute
801
21.02
40
Baseball (2.88 in diameter)
1.42
4.19x10 -3
121
Golf ball (1.65 in diameter)
0.5
1.40x10 -3
116
Raindrop (0.16 in diameter)
3.34X10 -4
1.29x10 -5
32
Although we’ve talked mostly about spheres in this section, the discussions on fluid flow
generally apply to any object moving through a fluid. Of course, the more complex the
object’s geometry, the harder it is to analyze the drag forces on it. Other factors such as
surface condition, and whether or not the object is at the interface between two fluids
(such as a ship in the ocean) further complicate the analysis. In practice, scale model
tests are particularly useful. In the Bibliography, we give several sources where you can
find more practical drag data for objects other than spheres.
Magnus Effect
The Magnus effect (also known as the Robbins effect) is quite an interesting phenom¬
enon. You know from the previous section that an object moving through a fluid en¬
counters drag. What would happen if that object were now spinning as it moved through
the fluid? For example, consider the sphere that we talked about earlier and assume that
while moving through a fluid such as air or water, it spins about an axis passing through
its center of mass. What happens when the sphere spins is the interesting part—it ac¬
tually generates lift! That’s right, lift. From everyday experience, most people usually
associate lift with a wing-like shape such as an airplane wing or a hydrofoil. It is far less
well known that cylinders and spheres can produce lift as well—that is, as long as they
are spinning. We’ll use the moving sphere to explain what’s happening here.
From the previous section on drag, you know that for a fast-moving sphere there will
be some point on the sphere where the flow separates, creating a turbulent wake behind
the sphere. Recall that the pressure acting on the sphere within this turbulent wake is
132 | Chapter 6: Projectiles
www.it-ebooks.info
lower than the pressure acting on the leading surface of the sphere, and this pressure
differential gives rise to the pressure drag component. When the sphere is spinning—
say, clockwise—about a horizontal axis passing through its center, as shown in
Figure 6-12, the fluid passing over the top of the sphere will be sped up while the fluid
passing under the sphere will be retarded.
Remember, because of friction, there is a thin boundary layer of fluid that attaches to
the sphere’s surface. At the sphere’s surface, the velocity of the fluid in the boundary
layer is 0 relative to the sphere. The velocity increases within the boundary layer as you
move further away from the sphere’s surface. In the case of the spinning sphere, there
is now a difference in fluid pressure above and below the sphere due to the increase in
velocity above the sphere and the decrease in velocity below the sphere. Further, the
separation point on the top side of the sphere will be pushed further back along the
sphere. The end result is an asymmetric flow pattern around the sphere with a net lift
force (due to the pressure differential) perpendicular to the direction of flow. If the
surface of the sphere is roughened a little, not only will frictional drag increase, but this
lift effect will increase as well.
Don’t let the term lift confuse you into thinking that this force always acts to lift, or
elevate, the sphere. The effect of this lift force on the sphere’s trajectory is very much
tied to the axis of rotation about which the sphere is spinning as related to the direction
in which the sphere is traveling (that is, its angular velocity).
The magnitude of the Magnus force is proportional to the speed of travel, rate of spin,
density of fluid, size of the object, and nature of the fluid flow. This force is not easy to
calculate analytically, and as with many problems in fluid dynamics, you must rely on
experimental data to accurately estimate it for a specific object under specific conditions.
There are, however, some analytical techniques that will allow you to approximate the
Magnus force. Without going into the theoretical details, you can apply the Kutta-
Magnus Effect | 133
www.it-ebooks.info
Joukouski theroem to estimate the lift force on rotating objects such as cylinders and
spheres. The Kutta-Joukouski theorem is based on a frictionless idealization of fluid
flow involving the concept of circulation around the object (like a vortex around the
object). You can find the details of this theory in any fluid dynamics text (we give some
references in the Bibliography), so we won’t go into the details here. However, we will
give you some results.
For a spinning circular cylinder moving through a fluid, you can use this formula to
estimate the Magnus lift force:
Fl = 2 k p L v r 2 a)
where v is the speed of travel, L is the length of the cylinder, r is its radius, and to is its
angular velocity in radians per second (rad/s). If you have spin, n, in revolutions per
second (rps), then to = 2 tt n. If you have spin, n, in revolutions per minute (rpm), then
to = (2 tt n) / 60.
For a spinning sphere moving through a fluid, you can use this formula:
F L = (2 ;t p v r 4 w) / (2 r)
where r is the radius of the sphere. Consistent units for these equations would yield lift
force in pounds in the English system or newtons in the SI system. In the SI system, the
appropriate units for these quantities are kg/m 3 , m/s, and m, respectively.
Keep in mind that these formulas only approximate the Magnus force; they’ll get you
in the ballpark, but they are not exact and actually could be off by up to 50% depending
on the situation. These formulas assume that 1) there is no slip between the fluid and
the rotating surface of the object, 2) there is no friction, 3) surface roughness is not taken
into account, and 4) there is no boundary layer.
At any rate, these equations will allow you to approximate the Magnus effect for flying
objects in your games, where you’ll be able to model the relative differences between
objects of different size that may be traveling at different speeds with different spin rates.
You’ll get the look right. If numerical accuracy is what you’re looking for, then you’ll
have to turn to experimental data for your specific problem.
Similar to the drag data shown in the previous section, experimental lift data is generally
presented in terms of lift coefficient. Using an equation similar to the drag equation,
you can calculate the lift force with the following equation:
F l = (0.5 p v 2 A) C L
As usual, it’s not as simple as this equation makes it appear. The trick is in determining
the lift coefficient, C L , which is a function of surface conditions, the Reynolds number,
134 | Chapter 6: Projectiles
www.it-ebooks.info
velocity, and spin rate. Further, experiments show that the drag coefficient is also affected
by spin.
For example, consider a golf ball struck perfectly (right!) such that the ball spins about
a horizontal axis perpendicular to its direction of travel while in flight. In this case the
Magnus force will tend to lift the ball higher in the air, increasing its flight time and
range. For a golf ball struck such that its initial velocity is 58 m/s with a take-off angle
of 10 degrees, the increase in range due to Magnus lift is on the order of 59 meters; thus,
it’s clear that this effect is significant. In fact, over the long history of the game of golf,
people have attempted to maximize this effect. In the late 1800s, when golf balls were
still made with smooth surfaces, players observed that used balls with roughened sur¬
faces flew even better than smooth balls. This observation prompted manufacturers to
start making balls with rough surfaces so as to maximize the Magnus lift effect. The
dimples that you see on modern golf balls are the result of many decades of experience
and research and are thought to be optimum.
Typically a golf ball takes off from the club with an initial velocity on the order of 76 m/
s, with a backspin on the order of 60 revolutions per second (rps). For these initial
conditions, the corresponding Magnus lift coefficient is within the range of 0.1 to 0.35.
Depending on the spin rate, this lift coefficient can be as high as 0.45, and the lift force
acting on the ball can be as much as 50% of the ball’s weight.
If the golf ball is struck with a less-than-perfect stroke (that’s more like it), the Magnus
lift force may work against you. For example, if your swing is such that the ball leaves
the club head spinning about an axis that is not horizontal, then the ball’s trajectory will
curve, resulting in a slice or a draw. If you top the ball such that the upper surface of the
ball is spinning away from you, then the ball will tend to curve downward much more
rapidly, significantly reducing the range of your shot.
As another example, consider a baseball pitched such that it’s spinning with topspin
about a horizontal axis perpendicular to its direction of travel. Here the Magnus force
will tend to cause the ball to curve in a downward direction, making it drop more rapidly
than it otherwise would without spin. If the pitcher spins the ball such that the axis of
rotation is not horizontal, then the ball will curve out of the vertical plane. Another trick
that pitchers use is to give the ball backspin, making it appear (to the batter) to actually
rise. This rising fastball does not actually rise, but because of the Magnus lift force it
falls much less rapidly than it would without spin.
For a typical pitched speed and spin rate of 45 m/s and 30 rps, respectively, the lift force
can be up to 33% of the ball’s weight. For a typical curveball, the lift coefficient is within
the range of 0.1 to 0.2, and for flyballs it can be up to 0.4.
These are only two examples; however, you need not look far to find other examples of
the Magnus force in action. Think about the behavior of cricket balls, soccer balls, tennis
balls, or ping-pong balls when they spin in flight. Bullets fired from a gun with a rifling
Magnus Effect | 135
www.it-ebooks.info
barrel also spin and are affected by this Magnus force. There have even been sailboats
built with tall, vertical, rotating cylindrical “sails” that use the Magnus force for pro¬
pulsion. We’ve also seen technical articles describing a propeller with spinning cylin¬
drical blades instead of airfoil-type blades.
To further illustrate the Magnus effect, we’ve prepared a simple example program that
simulates a ball being thrown with varying amounts of backspin (or topspin). This
example is based on the cannon example, so here again, the code should look familiar
to you. In this example we’ve neglected drag, so the only forces that the ball will see are
due to gravity and the Magnus effect. We did this to isolate the lift-generating effect of
spin and to keep the equations of motion clearer.
Since most of the code for this example is identical, or very similar, to the previous
cannon examples, we won’t repeat it here. We will, however, show you the global vari¬
ables used in this simulation along with a revised DoSi.mutation function that takes care
of the equations of motion:
//.//
// Global variables required for this simulation
//.//
TVector VI; // Initial velocity (given), m/s
TVector V2; // Velocity vector at time t, m/s
double m; // Projectile mass (given), kg
TVector si; // Initial position (given), m
TVector s2; // The projectile's position (displacement) vector, m
double time; // The time from the instant the projectile
// is launched, s
double tine; // The time increment to use when stepping
// through the simulation, s
double g; // acceleration due to gravity (given), m/s A 2
double spin; // spin in rpm (given)
double omega; // spin in radians per second
double radius; // radius of projectile (given), m
#define PI 3.14159f
#define RHO 1.225f // kg/m A 3
//.//
int DoSimulation(void)
//.....//
{
double C = PI * PI * RHO * radius * radius * radius * omega;
double t;
// step to the next time in the simulation
time+=tlnc;
t = time;
// Calc. V2:
V2.i = 1.0f/(1.0f-(t/m)*(t/m)*C*C) * (VI.i + C * VI.j * (t/m) -
136 | Chapter 6: Projectiles
www.it-ebooks.info
C * g * (t*t)/m);
V2.j = VI.j + (t/m)*C*V2.i - g*t;
// Calc. S2:
s2.1 = sl.i + VI.i * t + (1.0f/2.0f) * (C/n * V2.j) * (t*t);
s2.j = sl.j + VI.j * t + (1.0f/2.0f) * ( ((C*V2. i) - m*g)/m ) * (t*t);
// Check for collision with ground (x-z plane)
if(s2. j <= 0)
return 2;
// Cut off the simulation if it's taking too long
// This is so the program does not get stuck in the while loop
if(time>60)
return 3;
return 0;
}
The heart of this simulation is the lines that calculate V2 and s2, the instantaneous
velocity and position of the projectile, respectively. The equations of motion here come
from the 2D kinetic equations of motion including gravity, as discussed in Chapter 4,
combined with the following formula (shown earlier) for estimating the Magnus lift on
a spinning sphere:
F l = (2 n 2 p v r 4 a) / (2 r)
You can see the effect of spin on the projectile’s trajectory by providing the sample
program with different values for spin in revolutions per minute. The program converts
this to radians per second and stores this value in the variable omega. A positive spin
value indicates bottom spin such that the bottom of the sphere is spinning away from
you, while a negative spin indicates topspin, where the top of the ball spins away from
you. Bottom spin generates a positive lift force that will tend to extend the range of the
projectile, while topspin generates negative lift that will force the projectile toward the
ground, shortening its range. (Note that this example assumes that the spin axis is hor¬
izontal and perpendicular to the plane of the screen.) Figure 6-13 illustrates this be¬
havior.
Magnus Effect | 137
www.it-ebooks.info
Figure 6-13. Magnus effect sample program
Variable Mass
In Chapter 1 we mentioned that some problems in dynamics involve variable mass.
We’ll look at variable mass here since it applies to self-propelled projectiles such as
rockets. When a rocket is producing thrust to accelerate, it loses mass (fuel) at some
rate. When all of the fuel is consumed (burnout), the rocket no longer produces thrust
and has reached its maximum speed. After burnout you can treat the trajectory of the
rocket just as you would a non-self-propelled projectile, as discussed earlier. However,
while the rocket is producing thrust, you need to consider its mass change since this
will affect its motion.
In cases where the mass change of the object under consideration is such that the mass
being expelled or taken in has 0 absolute velocity—like a ship consuming fuel, for ex¬
ample—you can set up the equations of motion as you normally would, where the sum
of the forces equals the rate of change in momentum. However, in this case mass will
be a function of time, and your equations of motion will look like this:
EF = m a = d/dt (m v) =m (dv/dt) + (dm/dt) v
You can proceed to solve them just as you would normally, but keep in mind the time
dependence of mass.
A rocket, on the other hand, expels mass at some nonzero velocity, and you can’t use
the preceding approach to properly account for its mass change. In this case, you need
to consider the relative velocity between the expelled mass and the rocket itself. The
linear equation of motion now looks like this:
ZF = m dv/dt + dm/dt u
138 | Chapter 6: Projectiles
www.it-ebooks.info
where u is the relative velocity between the expelled mass and the object (the rocket in
this case).
For a rocket traveling straight up, neglecting air resistance and the pressure at the ex¬
haust nozzle, the only force acting on the rocket is due to gravity. But the rocket is
expelling mass (burning fuel). How it expels this mass is not important here, since the
forces involved are internal to the rocket; we need to consider only the external forces.
Let the fuel burn rate be -m’. The equation of motion (in the vertical direction) for the
rocket is as follows:
XF = m dv/dt + dm/dt u
-mg = m dv/dt - m’ u
If you rearrange this so that it looks like there’s only an ma term on the right of this
equation, you get:
m’ u - mg = m dv/dt = ma
Here you can see that the thrust that propels the rocket into the air is equal to m ’u. Since
the fuel burn rate is constant, the mass of the rocket at any instant in time is equal to:
m = m 0 - m’t
where m D is the initial mass, and the burn rate, m\ is in the form mass per unit time.
Variable Mass | 139
www.it-ebooks.info
www.it-ebooks.info
PART II
Rigid-Body Dynamics
Part II focuses on rigid-body dynamics and development of both single- and multibody
simulations. This part covers numerical integration, real-time simulation of particles
and rigid bodies, and connected rigid bodies. Generally, this part covers what most game
programmers consider elements of a physics engine.
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 7
Real-Time Simulations
This chapter is the first in a series of chapters designed to give you a thorough intro¬
duction to the subject of real-time simulation. We say introduction because the subject
is too vast and complex to adequately treat in a few chapters; however, we say thor¬
ough because we’ll do more than touch on real-time simulations. In fact, we’ll walk you
through the development of two simple simulations, one in two dimensions and the
other in three dimensions.
What we hope to do is give you enough of an understanding of this subject so that you
can pursue it further with confidence. In other words, we want you to have a solid
understanding of the fundamentals before jumping in to use someone else’s physics
engine, or venturing out to write your own.
In the context of this book, a real-time simulation is a process whereby you calculate
the state of the object (or objects) you’re trying to represent on the fly. You don’t rely on
prescripted motion sequences to animate your object, but instead you rely on your
physics model, the equations of motion, and your differential equation solver to take
care of the motion of your object as the simulation progresses. This sort of simulation
can be used to model rigid bodies like the airplane in our FlightSim example, or flexible
bodies such as cloth and human figures. Perhaps one of the most fundamental aspects
of implementing a real-time rigid-body simulator is solving the equations of motion
using numerical integration techniques. For this reason, we’ll spend this entire chapter
explaining the numerical integration techniques that you’ll use later in the 2D and 3D
simulators that we’ll develop.
If you refer back for a moment to Chapter 4, where we outlined a generic procedure for
solving kinetics problems, you’ll see that we’ve covered a lot of ground so far. The pre¬
ceding chapters have shown you how to estimate mass properties and develop the gov¬
erning equations of motion. This chapter will show you how to solve the equations of
motion in order to determine acceleration, velocity, and displacement. We’ll follow this
143
www.it-ebooks.info
chapter up with several showing you how to implement both 2D and 3D rigid-body
simulations.
Integrating the Equations of Motion
By now you should have a thorough understanding of the dynamic equations of motion
for particles and rigid bodies. If not, you may want to go back and review Chapter 1
through Chapter 4 before reading this one. The next aspect of dealing with the equations
of motion is actually solving them in your simulation. The equations of motion that
we’ve been discussing can be classified as ordinary differential equations. In Chap¬
ter 2 and Chapter 4, you were able to solve these differential equations explicitly since
you were dealing with simple functions for acceleration, velocity, and displacement.
This won’t be the case for your simulations. As you’ll see in later chapters, force and
moment calculations for your system can get pretty complicated and may even rely on
tabulated empirical data, which will prevent you from writing simple mathematical
functions that can be easily integrated. This means that you have to use numerical in¬
tegration techniques to approximately integrate the equations of motion. We say ap¬
proximately because solutions based on numerical integration won’t be exact and will
have a certain amount of error depending on the chosen method.
We’re going to start with a rather informal explanation of how we’ll apply numerical
integration because it will be easier to grasp. Later we’ll get into some of the formal
mathematics. Take a look at the differential equation of linear motion for a particle (or
rigid body’s center of mass):
F = m dv/dt
Recall that this equation is a statement of force equals mass times acceleration, where F
is force, m is mass, and dv/dt is the time derivative of velocity, which is acceleration. In
the simple examples of the earlier chapters of this book, we rewrote this equation in the
following form so it could be integrated explicitly:
dv/dt = F/m
dv = (F/m) dt
One way to interpret this equation is that an infinitesimally small change in velocity,
dv, is equal to (F/m) times an infinitesimally small change in time. In earlier examples,
we integrated explicitly by taking the definite integral of the left side of this equation
with respect to velocity and the right side with respect to time. In numerical integration
you have to take finite steps in time, thus dt goes from being infinitely small to some
discrete time increment, At, and you end up with a discrete change in velocity, Av:
Av = (F/m) At
144 | Chapter 7: Real-Time Simulations
www.it-ebooks.info
It is important to notice here that this does not give a formula for instantaneous velocity;
instead, it gives you only an approximation of the change in velocity. Thus, to approx¬
imate the actual velocity of your particle (or rigid body), you have to know what its
velocity was before the time change At. At the start of your simulation, at time 0, you
have to know the starting velocity of your particle. This is an initial condition and is
required in order to uniquely define your particle’s velocity as you step through time
using this equation: 1
Vt+At = v t + (F/m) At
where the initial condition is:
v t =o = v 0
Here v t is velocity at some time t, v t+At is velocity at some time plus the time step, At is
the time step, and v 0 is the initial velocity at time 0.
You can integrate the linear equation of motion one more time in order to approximate
your particles displacement (or position). Once you’ve determined the new velocity
value, at time t+ At, you can approximate displacement using:
S t+At — s t + At (v t+ At)
where the initial condition on displacement is:
S t =0 = So
The integration technique discussed here is known as Euler’s method, and it is the most
basic integration method. While Euler’s method is easy to grasp and fairly straightfor¬
ward to implement, it isn’t necessarily the most accurate method.
You can reason that the smaller you make your time step—that is, as At approaches dt
—the closer you’ll get to the exact solution. There are, however, computational problems
associated with using very small time steps. Specifically, you’ll need a huge number of
calculations at very small Af’s, and since your calculations won’t be exact (depending
on numerical precision you’ll be rounding off and truncating numbers), you’ll end up
with a buildup of round-off error. This means that there is a practical limit as to how
small a time step you can take. Fortunately, there are several numerical integration
techniques at your disposal that are designed to increase accuracy for reasonable step
sizes.
1. In mathematics, this sort of problem is termed an initial value problem.
Integrating the Equations of Motion | 145
www.it-ebooks.info
Even though we used the linear equation of motion for a particle, this integration tech¬
nique (and the ones we’ll show you later) applies equally well to the angular equations
of motion.
Euler's Method
The preceding explanation of Euler’s method was, as we said, informal. To treat Euler’s
method in a more mathematically rigorous way, we’ll look at the Taylor series expansion
of a general function, y{x ). Taylor’s theorem lets you approximate the value of a function
at some point by knowing something about that function and its derivatives at some
other point. This approximation is expressed as an infinite polynomial series of the
form:
y(x + Ax) = y(x) + (Ax) y'(x) + ((Ax) 2 / 2!) y"(x) + ((Ax) 3 / 3!)
y'"(x) + ■ ■ ■
where y is some function of x, (x + Ax) is the new value of x at which you want to
approximate y, y is the first derivative of y, y" is the second derivative of y, and so on.
In the case of the equation of motion discussed in the preceding section, the function
that you are trying to approximate is the velocity as a function of time. Thus, you can
write v(f) instead ofy(x), which yields the Taylor expansion:
v(t + At) = v(t) + (At) v'(t) + ((At) 2 / 2!) v"(t) + ((At) 3 / 3!) v"'(t)
+ • • •
Note here that v'(t) is equal to dv/dt, which equals F/m in the example equation of motion
discussed in the preceding section. Note also that you know the value of v at time t.
What you want to find is the value of v at time t + At knowing v at time t and its derivative
at time t. As a first approximation, and since you don’t know anything about v’s second,
third, or higher derivatives, you can truncate the polynomial series after the term (At)
v'(t), which yields:
v(t + At) = v(t) + (At) v'(t)
This is the Euler integration formula that you saw in the last section. Since Euler’s for¬
mula goes out only to the term that includes the first derivative, the rest of the series
that was left off is the truncation error. These terms that were left off are called higher-
order terms, and getting rid of them results in a first-order approximation. The rationale
behind this approximation is that the further you go in the series, the smaller the terms
and the less influence they have on the approximation. Since At is presumed to be a
small number, At 2 is even smaller, At 3 even smaller, and so on, and since these At terms
appear in the numerators, each successively higher-order term gets smaller and smaller.
146 | Chapter 7: Real-Time Simulations
www.it-ebooks.info
In this case, the first truncated term, ((At) 2 / 2!) v"(f), dominates the truncation error,
and this method is said to have an error of order (At) 2 .
Geometrically, Euler’s method approximates a new value, at the current step, for the
function under consideration by extrapolating in the direction of the derivative of the
function at the previous step. This is illustrated in Figure 7-1.
Figure 7-1. Euler integration step
Figure 7-1 illustrates the truncation error and shows that Eulers method will result in
a polygonal approximation of the smooth function under consideration. Clearly, if you
decrease the step size, you increase the number of polygonal segments and better ap¬
proximate the function. As we said before, though, this isn’t always efficient to do since
the number of computations in your simulation will increase and round-off error will
accumulate more rapidly.
To illustrate Euler’s method in practice, let’s examine the linear equation of motion for
the ship example of Chapter 4:
T - (C v) = ma
where T is the propeller’s thrust, C is a drag coefficient, v is the ship’s velocity, m its mass,
and a its acceleration.
Figure 7-2 shows the Euler integration solution, using a 0.5s time step, superimposed
over the exact solution derived in Chapter 4 for the ship’s speed over time.
Euler's Method | 147
www.it-ebooks.info
Zooming in on this graph allows you to see the error in the Euler approximation. This
is shown in Figure 7-3.
6.5 7 7.5 8 8.5
Time
Figure 7-3. Euler error
Table 7-1 shows the numerical values of speed versus time for the range shown in
Figure 7-3. Also shown in Table 7-1 is the percent difference, the error, between the
exact solution and the Euler solution at each time step.
148 | Chapter 7: Real-Time Simulations
www.it-ebooks.info
Table 7-1. Exact solution versus Euler solution
1 Time(s)
Velocity, exact (m/s)
Velocity, Euler (m/s)
Error 1
6.5
9.559084
9.733158
1.82%
7
10.06829
10.2465
1.77%
7.5
10.55267
10.73418
1.72%
8
11.01342
11.19747
1.67%
8.5
11.4517
11.63759
1.62%
As you can see, the truncation error in this example isn’t too bad. It could be better,
though, and we’ll show you some more accurate methods in a moment. Before that,
however, you should notice that in this example Euler’s method is also stable—that is,
it converges well with the exact solution as shown in Figure 7-4, where we’ve carried
the time range out further.
Here’s a code snippet that implements Euler’s method for this example:
// Global Variables
float
float
float
float
float
// thrust
// drag coefficient
// velocity
// mass
// displacement
// This function progresses the simulation by dt seconds using
Euler's Method | 149
www.it-ebooks.info
// Euler's basic method
void StepSimuiation(float dt)
{
float F; // total force
float A; // acceleration
float Vnew; // new velocity at time t + dt
float Snew; // new position at time t + dt
// Calculate the total force
F = (T - (C * V));
// Calculate the acceleration
A = F / M;
// Caiculate the new velocity at time t + dt
// where V is the velocity at time t
Vnew = V + A * dt;
// Calculate the new displacement at time t + dt
// where S is the displacement at time t
Snew = S + Vnew * dt;
// Update old velocity and displacement with the new ones
V = Vnew;
S = Snew;
}
Although Euler’s method is stable in this example, that isn’t always so, depending on the
problem you’re trying to solve. This is something that you must keep in mind when
implementing any numerical integration scheme. What we mean by stable here is that,
in this case, the Euler solution converges with the exact solution. An unstable solution
could manifest errors in two ways. First, successive values could oscillate above and
below the exact solution, never quite converging on it. Second, successive values could
diverge from the exact solution, creating a greater and greater error over time.
Take a look at Figure 7-5. This figure shows how Euler’s method can become very un¬
stable. What you see in the graph represents the vibratory motion of a spring-mass
system. This is a simple dynamical system that should exhibit regular sinusoidal motion.
150 | Chapter 7; Real-Time Simulations
www.it-ebooks.info
It’s clear from the figure that using Euler’s method yields terribly unstable results. You
can see how the motion amplitude continues to grow. If this were a game, say, where
you have a few objects connected by springs interacting with one another, then this sort
of instability would manifest itself by wildly unrealistic motion of those objects. Worse
yet, the simulation could blow up, numerically speaking.
Often, your choice of step size affects stability where smaller step sizes tend to eliminate
or minimize instability and larger steps aggravate the problem. If you’re working with
a particularly unwieldy function, you might find that you have to decrease your step
size substantially in order to achieve stability. This, however, increases the number of
computations you need to make. One way around this difficulty is to employ what’s
called an adaptive step size method, in which you change your step size on the fly de¬
pending on the magnitude of a predicted amount of truncation error from one step to
the next. If the truncation error is too large, then you back up a step, decrease your step
size, and try again.
One way to implement this for Euler’s method is to first take a step of size At to obtain
an estimate at time t + At, and then take two steps (starting from time t again) of size
At/2 to obtain another estimate at time t + At. Since we’ve been talking velocity in the
Euler's Method | 151
www.it-ebooks.info
examples so far, let’s call the first estimate v 1 and the second estimate v 2 . 2 A measure of
the truncation error is then:
e t = Ivi - v 2 l
If you wish to keep the truncation error within a specified limit, e to , then you can use
the following formula to find out what your step size should be in order to maintain the
desired accuracy:
Atnew — Atold ( e to I e t)
Here, Af old is the old time step and Af new is the new one that you should use to maintain
the desired accuracy. You’ll have to make this check for each time step, and if you find
that the error warrants a smaller time step, then you’ll have to back up a step and repeat
it with the new time step.
Here’s a revised StepSimulation function that implements this adaptive step size tech¬
nique, checking the truncation error on the velocity integration:
// New global variable
float eto; // truncation error tolerance
// This function progresses the simulation by dt seconds using
// Euler's basic method with an adaptive step size
void StepSimulation(float dt)
{
float
F;
// total force
float
A;
// acceleration
float
Vnew;
// new velocity at time
t + dt
float
Snew;
// new position at time
t + dt
float
VI, M2:
; // temporary velocity '
variables
float
dtnew;
// new time step
float
et;
// truncation error
// Take one step of size dt to estimate the new velocity
F = (T - (C * V));
A = F / M;
VI = V + A * dt;
// Take two steps of size dt/2 to estimate the new velocity
F = (T - (C * V));
A = F / M;
V2 = V + A * (dt/2);
F = (T - (C * M2))-,
2. Even though we’re talking about velocity and time here, these techniques apply to any function—for example,
displacement versus time, etc.
152 | Chapter 7: Real-Time Simulations
www.it-ebooks.info
A = F / M;
V2 = V2 + A * (dt/2);
// Estimate the truncation error
et = absf(Vl - V2);
// Estimate a new step size
dtnew = dt * SQRT(eto/et);
if (dtnew < dt)
{ // take at step at the new smaller step size
F = (T - (C * V));
A = F / M;
Vnew = V + A * dtnew;
Snew = S + Vnew * dtnew;
} else
{ // original step size is okay
Vnew = VI;
Snew = S + Vnew * dt;
}
// Update old velocity and displacement with the new ones
V = Vnew;
S = Snew;
}
Better Methods
At this point, you might be wondering why you can’t simply use more terms in the Taylor
series to reduce the truncation error of Euler’s method. In fact, this is the basis for several
integration methods that offer greater accuracy than Euler’s basic method for a given
step size. Part of the difficulty associated with picking up more terms in the Taylor’s
series expansion is in being able to determine the second, third, fourth, and higher
derivatives of the function you’re trying to integrate. The way around this problem is
to perform additional Taylor series expansions to approximate the derivatives of the
function under consideration and then substitute those values back into your original
expansion.
Taking this approach to include one more Taylor term beyond the basic Euler method
yields a so-called improved Euler method that has a reduced truncation error on the
order of (At ) 3 instead of (At) 2 . The formulas for this method are:
kt = (At) y'(t, y)
k 2 = (At) y'(t + At, y + k;)
y(t + At) = y(t) + Yi (k! + k 2 )
Better Methods | 153
www.it-ebooks.info
Here y is a function of t, and y is the derivative as a function of t and possibly of y
depending on the equations you’re trying to solve, and At the step size.
To make this clearer for you, we’ll show these formulas in terms of the ship example
equation of motion of Chapter 4, the same example that we discussed in the previous
section. In this case, velocity is approximated by the following formulas:
k! = At [1/m (T - C v t )]
k 2 = AUl/mCT-CCvt + k!))]
v t +At = V t + V4(ki + k 2 )
where v t is the velocity at time t, and v t+At is the new velocity at time t+At.
Here is the revised StepSimulation function showing how to implement this method
in code:
// This function progresses the simulation by dt seconds using
// the "improved" Euler method
void StepSimulation(float dt)
{
float F; // total force
float A; // acceleration
float Vnew; // new velocity at time t + dt
float Snew; // new position at time t + dt
float kl, k2;
F = (T - (C * V));
A = F/M;
kl = dt * A;
F = (T - (C * (V + kl)));
A = F/M;
k2 = dt * A;
// Calculate the new velocity at time t + dt
// where V is the velocity at time t
Vnew = V + (kl + k2) / 2;
// Calculate the new displacement at time t + dt
// where S is the displacement at time t
Snew = S + Vnew * dt;
// Update old velocity and displacement with the new ones
V = Vnew;
S = Snew;
}
154 | Chapter 7: Real-Time Simulations
www.it-ebooks.info
We can carry out this procedure of taking on more Taylor terms even further. The
popular Runge-Kutta method takes such an approach to reduce the truncation error to
the order of (At) 5 . The integration formulas for this method are as follows:
kt = (At) y'(t, y)
k 2 = (At) y'(t + At/2, y + k x /2)
k 3 = (At) y'(t + At/2, y + k 2 /2)
k 4 = (At) y'(t + At, y + k 3 )
y(t + At) = y(t) + 1/6 (k 3 + 2 (k 2 ) + 2 (k 3 ) + k 4 )
Applying these formulas to our ship example yields:
k, = At [1/m (T - C v t )]
k 2 = At [1/m (T - C (v t + k 3 /2))]
k 3 = At [1/m (T - C (v t + k 2 /2))]
k 4 = At[l/m(T-C(v t + k 3 ))]
V t+At — v t + 1/6 (kj + 2 (k 2 ) + 2 (k 3 ) + IC 4 )
For our example, the Runge-Kutta method is implemented as follows:
// This function progresses the simulation by dt seconds using
// the Runge-Kutta method
void StepSimulation(float dt)
{
float F; // total force
float A; // acceleration
float Vnew; // new velocity at time t + dt
float Snew; // new position at time t + dt
float kl, k2, k3, k4;
F = (T - (C * V));
A = F/M;
kl = dt * A;
F = (T - (C * (V + kl/2)));
A = F/M;
k2 = dt * A;
F = (T - (C * (V + k2/2)));
A = F/M;
k3 = dt * A;
F = (T - (C * (V + k3)));
A = F/M;
k4 = dt * A;
Better Methods |
155
www.it-ebooks.info
// Calculate the new velocity at tine t + dt
// where V is the velocity at tine t
Vnew = V + (kl + 2*k2 + 2*k3 + k4) / 6;
// Calculate the new displacenent at tine t + dt
// where S is the displacenent at tine t
Snew = S + Vnew * dt;
// Update old velocity and displacenent with the new ones
V = Vnew;
S = Snew;
}
To show you how accuracy is improved over the basic Euler method, we’ve superim¬
posed integration results for the ship example using these two methods over those
shown in Figure 7-2 and Figure 7-3. Figure 7-6 and Figure 7-7 show the results, where
Figure 7-7 is a zoomed view of 7-6.
156 | Chapter 7: Real-Time Simulations
www.it-ebooks.info
11.5
11
I 10.5
£
10
9.5
6.5 7 7.5 8 8.5
Time
Figure 7-7. A closer look
As you can see from these figures, it’s impossible to discern the curves for the improved
Euler and Runge-Kutta methods from the exact solution because they fall almost right
on top of each other. These results clearly show the improvement in accuracy over the
basic Euler method, whose curve is distinct from the other three. Over the interval from
6.5 to 8.5 seconds, the average truncation error is 1.72%, 0.03%, and 3.6xl0 _6 % for
Euler’s method, the improved Euler method, the Runge-Kutta method, respectively. It
is obvious, based on these results, that for this problem, the Runge-Kutta method yields
substantially better results for a given step size than the other two methods. Of course,
you pay for this accuracy, since you have several more computations per step in the
Runge-Kutta method.
Both of these methods are generally more stable than Euler’s method, which is a huge
benefit in real-time applications. Recall our discussion earlier about the stability of Eu¬
ler’s method. Figure 7-5 showed the results of applying Euler’s method to an oscillating
dynamical system. There, the motion results that should be sinusoidal were wildly er¬
ratic (i.e., unstable). Applying the improved Euler method, or the Runge-Kutta method,
to the same problem yields stable results, as shown in Figure 7-8.
Exact
Euler
Improved
R-K
Better Methods | 157
www.it-ebooks.info
Figure 7-8. Stable results using the improved Euler or the Runge-Kutta methods
Here the oscillatory motion is clearly sinusoidal, as it should be. The results for this
particular problem are almost identical whether you use the improved Euler method or
the Runge-Kutta method. Since for this problem the results of both methods are virtually
the same, you can save computational time and memory using the improved Euler
method versus the Runge-Kutta method. This can be a significant advantage for real¬
time games. Remember the Runge-Kutta method requires four derivative computations
per time step.
These methods aren’t the only ones at your disposal, but they are the most common.
The Runge-Kutta method is particularly popular as a general-purpose numerical inte¬
gration scheme. Other methods attempt to improve computational efficiency even fur¬
ther—that is, they are designed to minimize truncation error while still allowing you to
take relatively large step sizes so as to reduce the number of steps you have to take in
your integration. Still other methods are especially tailored for specific problem types.
We cite some pretty good references for further reading on this subject in the Bibliog¬
raphy.
158 | Chapter 7: Real-Time Simulations
www.it-ebooks.info
Summary
At this point you should be comfortable with the terms that appear in the equations of
motion and be able to calculate terms like the sum of forces and moments, mass, and
inertia. You should also have a solid understanding of basic numerical integration tech¬
niques. Implementing these techniques in code is really straightforward since they are
composed of simple polynomial functions. The hard part is developing the derivative
function for your problem. In the case of the equations of motion, the derivative function
will include all your force and moment calculations for the particle or rigid body that
you are modeling. You’ll see some more numerical integration code when you get to
Chapter 8, Chapter 9, and Chapter 11.
Summary | 159
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 8
Particles
In this chapter we’ll show you how to apply what you’ve learned in Chapter 7 in a simple
particle simulator. Before getting to the specifics of the example we’ll present, let’s con¬
sider particles in general. Particles are simple idealizations that can be used to simulate
all sorts of phenomena or special effects within a game. For example, particle simulations
are often used to simulate smoke, fire, and explosions. They can also be used to simulate
water, dust clouds, and swarms of insects, among many other things. Really, your imag¬
ination is the only limit. Particles lend themselves to simulating both discrete objects
like bouncing balls and continua like water. Plus, you can easily ascribe an array of
attributes to particles depending on what you’re modeling.
For example, say, you’re modeling fire using particles. Each particle will rise in the air,
and as it cools its color will change until it fades away. You can tie the particle’s color to
its temperature, which is modeled using thermodynamics. The attribute you’d want to
track is the particle’s temperature. In a previous work, AI for Game Programmers
(O’Reilly), this book’s coauthor David M. Bourg used particles to represent swarms of
insects that would swarm, flock, chase, and evade depending on the artificial intelligence
(AI). The AI controlled their behavior, which was then implemented as a system of
particles using principles very similar to what you’ll see in this chapter.
Particles are not limited to collections of independent objects either. Later in this book,
you’ll learn how to connect particles together using springs to create deformable objects
such as cloth. Particles are extremely versatile, and you’ll do well to learn how to leverage
their simplicity.
You can use particles to model sand in a simple phone application that simulates an
hourglass. Couple this sand model with the accelerometer techniques you’ll learn about
in Chapter 21, and you’ll be able to make the sand flow by turning the phone over.
You can easily use particles to simulate bullets flying out of a gun. Imagine a Gatling
gun spewing forth a hail of lead, all simulated using simple particles. Speaking of spew-
161
www.it-ebooks.info
ing, how about using particles to simulate debris flung from an erupting volcano as a
special effect in your adventure game set in prehistoric times? Remember the Wooly
Willy toy? To make particles a direct part of game play, consider a diversionary appli¬
cation where you drag piles of virtual magnetic particles around a portrait photograph,
giving someone a lovely beard or mustache much like Wooly Willy.
Hopefully, you’re now thinking of creative ways to use particles in your games. So, let’s
address implementation. There are two basic ingredients to implementing a particle
simulator: the particle model and the integrator. (Well, you could argue that a third basic
ingredient is the renderer, where you actually draw the particles, but that’s more graphics
than physics, and we’re focusing on modeling and integrating in this book.)
The model very simply describes the attributes of the particles in the simulation along
with their rules of behavior. We mean this in the physics sense and not the AI sense in
this book, although in general the model you implement may very well include suitable
AI rules. Now, the integrator is responsible for updating the state of the particles
throughout the simulation. In this chapter, the particles’ states will be described by their
position and velocity at any given time. The integrator will update each particle’s state
under the influence of several external stimuli—forces such as gravity, aerodynamic
drag, and collisions.
The rest of this chapter will walk you through the details of a simple particle simulation
in an incremental manner. The first task willbe to simulate a set of particles falling under
the influence of gravity alone. Even though this sounds elementary, such an example
encompasses all of the basic ingredients mentioned earlier. Once gravity is under con¬
trol, we’ll show you how to implement still air drag and wind forces to influence the
particles’ motion. Then, we’ll make things more interesting by showing you how to
implement collision response between the particles and a ground plane plus random
obstacles. This collision stuff will draw on material presented in Chapter 5, so be sure
to read that chapter first if you have not already done so.
Figure 8-1 through Figure 8-4 show a few frames of this example simulation complete
with obstacles and collisions. Use your imagination here to visualize the particles falling
under the influence of gravity until they impact the circular objects, at which time they
bounce around and ultimately settle on the ground.
162 | Chapter 8: Particles
www.it-ebooks.info
Figure 8-1. Particles falling under the influence of gravity
Particles | 163
www.it-ebooks.info
164 | Chapter 8: Particles
www.it-ebooks.info
Figure 8-3. More collisions
Particles | 165
www.it-ebooks.info
Figure 8-4. Particles coming to rest on the ground plane
While working through this chapter, keep in mind that everything you’re learning here
will be directly applicable to 2D and 3D simulations. Chapters following this one will
build on the material covered here. We’ll focus on two dimensions in this chapter and
later in the book we’ll show you how to extend the simulation to 3D. Actually, for particle
simulations it’s almost trivial to make the leap from 2D to 3D. Trust us on this for now.
Simple Particle Model
The particle model we’ll begin with is very simple. All we want to achieve at first is to
have the particles fall under the influence of gravity. The particles will be initialized with
an altitude above a ground plane. Upon the start of the simulation, gravity will act on
each particle, continuously causing each to accelerate toward the ground plane, gaining
speed as it goes. Imagine holding a handful of small rocks up high and then releasing
them. Simple, huh?
166 | Chapter 8: Particles
www.it-ebooks.info
There are several particle attributes we must consider even for this simple example. Our
model assumes that each particle has mass, and a set diameter (we’re assuming our
particles are circles in 2D or spheres in 3D), occupies some position in space, and is
traveling at some velocity. Additionally, each particle is acted upon by some net external
force, which is the aggregate of all forces acting on the particle. These forces will be
gravity alone to start with, but will eventually include drag and impact forces. We set
up a Particle class to encapsulate these attributes as follows:
class Particle {
public:
float
fMass;
// Total mass
Vector
vPosltlon;
// Position
Vector
vVeloclty;
// Velocity
float
fSpeed;
// Speed (magnitude of the velocity)
Vector
vForces;
// Total force acting on the particle
float
fRadlus;
// Particle radius used for collision detection
Vector
vGravlty;
// Gravity force vector
Particle(void);
// Constructor
void
CalcLoads(vold); // Aggregates forces acting on the particle
void
UpdateBodyEuler(double dt); // Integrates one time step
void
Draw(vold);
// Draws the particle
};
Most of these attributes are self-explanatory given the comments we’ve included. Notice
that several of these attributes are Vecto r types. These vectors are defined in the custom
math library we’ve included in Appendix A. This type makes managing vectors and
performing arithmetic operations with them a breeze. Take a look at Appendix A to see
what this custom type does. We’ll just remind you of the data structure Vector uses:
three scalars called x, y, and z representing the three dimensions of a location or of a
movement in some direction. The z component will always be set to 0 in this chapter’s
examples.
You should have noticed the f Speed property in the Particle class. This property stores
the magnitude of the velocity vector, the particle’s speed. We’ll use this later when com¬
puting aerodynamic drag forces. We’ve also included a Vector type property called
vG r a vi ty, which stores the gravity force vector defining the magnitude and the direction
in which the gravity force acts. This is not really necessary, as you could hardcode the
gravity force vector or use a global variable; however, we’ve included it here to illustrate
some creative flexibility. For example, you could redefine the gravity vector in a game
that uses accelerometer input to determine gravity’s direction with respect to a particular
device’s orientation (see Chapter 21). And you may have a game where some particles
react to different gravities depending on their type, which can be of your own concoc¬
tion.
Simple Particle Model | 167
www.it-ebooks.info
Aside from properties, you’ll notice several methods in the Particle class. The con¬
structor is trivial. It sets everything to 0 except the particles mass, radius, and the gravity
force vector. The following code illustrates how everything is initialized:
Particle::Partlcle(vold)
{
fMass = 1.0;
vPosltion.x = 0.0;
vPosltion.y = 0.0;
vPositlon.z = 0.0;
vVelocity.x = 0.0;
vVelocity.y = 0.0;
vVelocity.z = 0.0;
fSpeed = 0.0;
vForces.x = 0.0;
vForces.y = 0.0;
vForces.z = 0.0;
fRadius = 0.1;
vGravlty.x = 0;
vGravtty.y = fMass * _GRAVITYACCELERATION;
}
Now is probably a good time to explain the coordinate system we’ve assumed. Our world
origin is located at the lower-left corner of the example program’s window with positive
x pointing to the right and positive y pointing up. The acceleration due to gravity acts
downward (i.e., in the negative y-direction). We’re using the SI system of units and have
defined the acceleration due to gravity as follows:
#deflne _GRAVITYACCELERATION -9.8f
That’s 9.8 m/s 2 in the negative y-direction. We’ve set the mass of each particle to 1 kg by
default, which means the force due to gravity is 1 kg times 9.8 m/s 2 , or 9.8 newtons of
force. We’ve set the radius of each particle to one-tenth of a meter. These masses and
radii are really arbitrary; you can set them to anything suitable for what you’re modeling.
The CalcLoads method is responsible for computing all the loads—forces—acting on
the particle, with the exception of impact forces (we’ll handle those later). For now, the
only force acting on the particles is that due to gravity, or simply, the weight of each
particle. CalcLoads is very simple at this point:
void Particle::CalcLoads(void)
{
// Reset forces:
vForces.x = 0.0f;
vForces.y = 0.0f;
// Aggregate forces:
vForces += vGravlty;
}
168 | Chapter 8: Partides
www.it-ebooks.info
The first order of business is to reset the vForces vector. vForces is the vector containing
the net force acting on the particle. All of these forces are aggregated in CalcLoads, as
shown by the line vForces += vGravity. Again, so far, the only force to aggregate is
that due to gravity.
Integrator
The UpdateBodyEuler method integrates the equations of motion for each particle.
Since we’re dealing with particles, the only equation of motion we need concern our¬
selves with is that for translation; rotation does not matter for particles (at least not for
us here). The following code sample shows UpdateBodyEuler.
void Particle::UpdateBodyEuler(double dt)
{
Vector a;
Vector dv;
Vector ds;
// Integrate equation of motion:
a = vForces / fMass;
dv = a * dt;
vVeloctty += dv;
ds = vVeloctty * dt;
vPosltlon += ds;
// Mtsc. calculations:
fSpeed = vVeloctty .MagnttudeQ;
}
As the name of this method implies, we’ve implemented Euler’s method of integration
as described in Chapter 7. Using this method, we simply need to divide the aggregate
forces acting on a particle by the mass of the particle to get the particle’s acceleration.
The line of code a = vForces / fMass does just this. Notice here that a is a Vector, as
is vForces. fMass is a scalar, and the / operator defined in the Vector class takes care
of dividing each component of the vForces vector by fMass and setting the corre¬
sponding components in a. The change in velocity, dv, is equal to acceleration times the
change in time, dt. The particle’s new velocity is then computed by the line vVeloctty
+= dv. Here again, vVeloctty and dv are Vectors and the += operator takes care of the
vector arithmetic. This is the first actual integration.
The second integration takes place in the next few lines, where we determine the par¬
ticle’s displacement and new position by integrating its velocity. The line ds = vVeloc
ity * dt determines the displacement, or change in the particle’s position, and the line
vPosltlon += ds computes the new position by adding the displacement to the parti¬
cle’s old position.
Simple Partide Model | 169
www.it-ebooks.info
The last line in UpdateBody Euler computes the particle’s speed by taking the magnitude
of its velocity vector.
For demonstration purposes, using Euler’s method is just fine. In an actual game, the
more robust method described in Chapter 7 is advised.
Rendering
In this example, rendering the particles is rather trivial. All we do is draw little circles
using Windows API calls wrapped in our own functions to hide some of the Windows-
specific code. The following code snippet is all we need to render the particles.
void Particle::Draw(void)
{
RECT r;
float drawRadius = max(2, fRadius);
SetRect(&r, vPosition.x - drawRadius,
_WINHEIGHT - (vPosition.y - drawRadius),
vPosition.x + drawRadius,
_WINHEIGHT - (vPosition.y + drawRadius));
DrawEllipse(&r, 2, RGB(0,0,0));
}
You can use your own rendering code here, of course, and all you really need to pay
close attention to is converting from world coordinates to window coordinates. Re¬
member, we’ve assumed our world coordinate system origin is in the lower-left corner
of the window, whereas the window drawing coordinate system has its origin in the
upper-left corner of the window. To transform coordinates in this example, all you need
to do is subtract the particle’s y-position from the height of the window.
The Basic Simulator
The heart of this simulation is handled by the Particle class described earlier. However,
we need to show you how that class is used in the context of the main program.
First, we define a few global variables as follows:
// Global Variables:
int FrameCounter = 0;
Particle Units[_MAX_NUM_UNITS];
FrameCounter counts the number of time steps integrated before the graphics display
is updated. How many time steps you allow the simulation to integrate before updating
the display is a matter of tuning. You’ll see how this is used momentarily when we discuss
the UpdateSimulation function. Units is an array of Particle types. These will rep¬
resent moving particles in the simulation—the ones that fall from above and bounce off
the circular objects we’ll add later.
170 | Chapter 8: Particles
www.it-ebooks.info
For the most part, each unit is initialized in accordance with the Particle constructor
shown earlier. However, their positions are all at the origin, so we make a call to the
following Initialize function to randomly distribute the particles in the upper-middle
portion of the screen within a rectangle of width _SPAWN_AREA_R*4 and a height of
_SPAWN_AREA_R, where _SPAWN_AREA_R is just a global define we made up.
boot Initialize(void)
{
int i;
GetRandomNumber(0, _WINWIDTH, true);
for(i=0; i<_MAX_N UM_U NIT S; 1++)
{
Unitsfi].vPosition.x = GetRandomNumber(_WINWIDTH/2-_SPAWN_AREA_R*2,
_WINWIDTH/2+_SPAWN_AREA_R*2, false);
Unitsfi].vPosition.y = _WINHEIGHT -
GetRandomNumber(_WINHEIGHT/2-_SPAWN_AREA_R,
_WINHEIGHT/2, false);
}
return true;
}
OK, now let’s consider UpdateSimulation as shown in the code snippet that follows.
This function gets called every cycle through the programs main message loop and is
responsible for cycling through all the Units, making appropriate function calls to up¬
date their positions, and rendering the scene.
void UpdateSimulation(void)
{
double dt = _TIMESTEP;
int i;
// initialize the back buffer
if(FrameCounter >= _RENDER_FRAME_COUNT)
{
ClearBackBuffer();
}
// update the particles (Units)
for(i=0; i<_MAX_NUM_UNITS; U+)
{
Unitsfi] .CalcLoadsQ;
Unitsfi].UpdateBodyEuler(dt);
if(FrameCounter >= _RENDER_FRAME_COUNT)
{
Units[i].Draw();
}
The Basic Simulator | 171
www.it-ebooks.info
lf(Units[i].vPosition.x > _WINWIDTH) Unitsft].vPosition.x = 0;
lf(Units[i].vPosition.x < 0) Unlts[l].vPosition.x = _WINWIDTH;
if(Units[i].vPosition.y > _WINHEIGHT) Units[i].vPosition.y = 0;
if(Units[i].vPosition.y < 0) Units[i].vPosition.y = _WINHEIGHT;
}
// Render the scene if required
if(FrameCounter >= _RENDER_FRAME_COUNT) {
CopyBackBufferToWindow();
FraneCounter = 0;
} eise
FrameCounter++;
}
The two local variables in UpdateSimulation are dt and i. i is trivial and serves as a
counter variable, dt represents the small yet finite amount of time, in seconds, over
which each integration step is taken. The global def ine_TIMESTEP stores the time step,
which we have set to 0.1 seconds. This value is subject to tuning, which we’ll discuss
toward the end of this chapter in the section “Tuning” on page 186.
The next segment of code checks the value of the frame counter, and if the frame counter
has reached the defined number of frames, stored in _RENDER_FRAME_COUNT, then the
back buffer is cleared to prepare it for drawing upon and ultimately copying to the
screen.
The next section of code under the comment update the particles does just that by
calling the CalcLoads and UpdateBodyEuler methods of each Unit. These two lines are
responsible for updating all the forces acting on each particle and then integrating the
equation of motion for each particle.
The next few lines within the for loop draw each particle if required and wrap each
particles position around the window extents should they progress beyond the edges
of the window. Note that we’re using window coordinates in this example.
Implementing External Forces
We’ll add a couple of simple external forces to start with—still air drag, and wind force.
We’ll use the formulas presented in Chapter 3 to approximate these forces, treating them
in a similar manner. Recall that still air drag is the aerodynamic drag force acting against
an object moving at some speed through still air. Drag always acts to resist motion.
While we’ll use the same formulas to compute a wind force, recall that wind force may
not necessarily act to impede motion. You could have a tailwind pushing an object along,
or the wind could come from any direction with components that push the object side¬
ways. In this example we’ll assume a side wind from left to right, acting to push the
particles sideways, with the still air drag resisting their falling motion. When we add
collisions later, this same drag formulation will act to resist particle motion in any di¬
rection in which they travel as they bounce around.
172 | Chapter 8: Particles
www.it-ebooks.info
The formula we’ll use to model still air drag is:
F d = >/ 2 pV 2 AC d
Here F d is the magnitude of the drag force. Its direction is directly opposite the velocity
of the moving particle, p is the density of air through which the particle moves, V is the
magnitude of the particles velocity (its speed), A is the projected area of the particle as
though it’s a sphere, and C d is a drag coefficient.
We can use this same formula to estimate the wind force pushing the particle sideways.
The only difference this time is that V is the wind speed, and the direction of the resulting
wind force is from our assumed left-to-right direction.
To add these two forces to our simulation, we need to make a few additions to the
Particle class’s CalcLoads method. The following code shows how CalcLoads looks
now. Remember, all we had in here originally were the first three lines of executable
code shown next—the code that initializes the aggregate force vector and then the line
of code that adds the force due to gravity to the aggregate force. All the rest of the code
in this method is new.
void Particle::CalcLoads(void)
{
// Reset forces:
vForces.x = 0.0f;
vForces.y = 0.0f;
// Aggregate forces:
// Gravity
vForces += vGravity;
// Still air drag
Vector vDrag;
Float fDrag;
vDrag-=vVelocity;
vDrag. NorpializeQ;
fDrag = 0.5 * _AIRDENSITY * fSpeed * fSpeed *
(3.14159 * fRadius * fRadius) * _DRAGCOEFFICIENT;
vDrag*=fDrag;
vForces += vDrag;
// Wind
Vector vWind;
vWind.x = 0.5 * _AIRDENSITY * _WINDSPEED *
_WINDSPEED * (3.14159 * fRadius * fRadius) *
_DRAGCOEFFICIENT;
Implementing External Forces | 173
www.it-ebooks.info
vForces += vklind;
}
So after the force due to gravity is added to the aggregate, two new local variables are
declared. vDrag is a vector that will represent the still air drag force. fDrag is the mag¬
nitude of that drag force. Since we know the drag force vector is exactly opposite to the
particle’s velocity vector, we can equate vDrag to negative vVelocity and then normalize
vDrag to obtain a unit vector pointing in a direction opposite of the particles velocity.
Next we compute the magnitude of the drag force using the formula shown earlier. This
line handles that:
fDrag = 0.5 * _AIRDENSITY * fSpeed * fSpeed *
(3.14159 * fRadius * fRadius) * _DRAGCOEFFICIENT;
Here, _AIRDENSITY is a global define representing the density of air, which we have set
to 1.23 kg/m 3 (standard air at 15°C). fSpeed is the particle’s speed: the magnitude of its
velocity. The 3.14159 * fRadius * fRadius line represents the projected area of the
particle assuming the particle is a sphere. And finally, _DRAGCOEFFICIENT is a drag co¬
efficient that we have set to 0.6. We picked this value from the chart of drag coefficient
for a smooth sphere versus the Reynolds number shown in Chapter 6. We simply eye-
balled a value in the Reynolds number range from le4 to le5. You have a choice here
of tuning the drag coefficient value to achieve some desired effect, or you can create a
curve fit or lookup table to select a drag coefficient corresponding to the Reynolds
number of the moving particle.
Now that we have the magnitude of the drag force, we simply multiply that force by the
unit drag vector to obtain the final drag force vector. This vector then gets aggregated
in vForces.
We handle the wind force in a similar manner with a few differences in the details. First,
since we know the unit wind force vector is in the positive x-direction (i.e., it acts from
left to right), we can simply set the x component of the wind force vector, vWind, to the
magnitude of the wind force. We compute that wind force using the same formula we
used for still air drag with the exception of using the wind speed instead of the particle’s
speed. We used _WINDSPEED, a global define, to represent the wind speed, which we
have set to 10 m/s (about 20 knots).
Finally, the wind force is aggregated in vForces.
At this stage the particles will fall under the influence of gravity, but not as fast as they
would have without the drag force. And now they’ll also drift to the right due to the
wind force.
174 | Chapter 8: Partides
www.it-ebooks.info
Implementing Collisions
Adding external forces made the simulation a little more interesting. However, to really
make it pop, we’re going to add collisions. Specifically, we’ll handle particle-to-ground
collisions and particle-to-object collisions. If you have not yet read Chapter 5, which
covers collisions, you should because we’ll implement principles covered in that chapter
here in the example. We’ll implement enough collision handling in this example to allow
particles to bounce off the ground and circular objects, and we’ll come back to collision
handling in more detail in Chapter 10. The material in this chapter should whet your
appetite. We’ll start with the easier case of particle-to-ground collisions.
Partide-to-Ground Collisions
Essentially what we’re aiming to achieve with particle-to-ground collision detection is
to prevent the particles from passing through a ground plane specified at some y coor¬
dinate. Imagine a horizontal impenetrable surface that the particles cannot pass
through. There are several things we must do in order to detect whether a particle is
indeed colliding with the ground plane. If so, then we need to handle the collision,
making the particles respond in a suitable manner.
The left side of Figure 8-5 illustrates a collision scenario. It’s easy to determine whether
or not a collision has taken place. Over a given simulation time step, a particle may have
moved from some previous position (its position at the previous time step) to its current
position. If this current position puts the centroid coordinate of the particle within one
particle radius of the ground plane, then a collision might be occurring. We say might
because the other criteria we need to check in order to determine whether or not a
collision is happening is whether or not the particle is moving toward the ground plane.
If the particle is moving toward the ground plane and it’s within one radius of the ground
plane, then a collision is occurring. It may also be the case that the particle has passed
completely through the ground plane, in which case we assume a collision has occurred.
Implementing Collisions | 175
www.it-ebooks.info
Previous Position
Collision Detected Penetration Prevented
Figure 8-5. Particle-to-ground collision
To prevent such penetration of the ground plane, we need to do two things. First, we
must reposition the particle so that it is just touching the ground plane, as shown on
the right side of Figure 8-5. Second, we must apply some impact force resulting from
the collision in order to force the particle to either stop moving down into the ground
plane or to move away from the ground plane. All these steps make up collision detection
and response.
There are several changes and additions that we must make to the example code in order
to implement particle-to-ground collision detection and response. Let’s begin with the
Particle class.
We’ve added three new properties to Particle, as follows:
class Particle {
Vector vPreviousPosition;
Vector vlmpactForces;
bool bCollision;
};
vPreviousPosition is used to store the particle’s position at the previous time step—
that is, at time t-dt. vlmpactForces is used to aggregate all of the impact forces acting
176 | Chapter 8: Particles
www.it-ebooks.info
on a particle. You’ll see later that it is possible for a particle to collide with more than
one object at the same time. bCollision is simply a flag that is used to indicate whether
or not a collision has been detected with the particle at the current time step. This is
important because when a collision occurs, at that instant in time, we assume that the
only forces acting on the particle are the impact forces; all of the other forces—gravity,
drag, and wind—are ignored for that time instant. We use bCollision in the updated
CalcLoads method:
void Particle::CalcLoads(void)
{
// Reset forces:
vForces.x = 0.0f;
vForces.y = 0.0f;
// Aggregate forces:
if(bCollision) {
// Add Impact forces (if any)
vForces += vlmpactForces;
} else {
// Gravity
vForces += vGravity;
// Still air drag
Vector vDrag;
float fDrag;
vDrag -= vVelocity;
vDrag.Normalize();
fDrag = 0.5 * _AIRDENSITY * fSpeed * fSpeed *
(3.14159 * fRadius * fRadius) * _DRAGCOEFFICIENT;
vDrag *= fDrag;
vForces += vDrag;
// Wind
Vector vWind;
vWind.x = 0.5 * _AIRDENSITY * _WINDSPEED * _WINDSPEED *
(3.14159 * fRadius * fRadius) * _DRAGCOEFFICIENT;
vForces += vWind;
}
}
The only difference between this version of CalcLoads and the previous one is that we
added the conditional if (bCollision) { ... } else { ... }. If bCollision is true,
then we have a collision to deal with and the only forces that get aggregated are the
impact forces. If there is no collision, if bCollision is false, then the non-impact forces
are aggregated in the usual manner.
You may have caught that we are aggregating impact forces in this example. This is an
alternate approach to the one shown in Chapter 5 . There we showed you how to calculate
an impulse and change an object s velocity in response to a collision, using conservation
Implementing Collisions | 177
www.it-ebooks.info
of momentum. Well, we’re still calculating impulses just like in Chapter 5; however, in
this example, we’re going to compute the impact force based on that impulse and apply
that force to the colliding objects. We’ll let the numerical integrator integrate that force
to derive the colliding particle’s new velocities. Either method works, and we’ll show
you an example of the former method later. We’re showing the latter method here just
to illustrate some alternatives. The advantage of this latter method is that it is easy to
compute impact forces due to multiple impacts and let the integrator take care of them
all at once.
Now, with these changes made to Particle, we need to add a line of code to Up
dateSimulation, as shown here:
void UpdateSimulation(vold)
1
// update computer controtled units:
for(i=0; i<_MAX_NUM_UNITS; i++)
{
Unitsfi].bColtision = CheckForCollisions(&(Units[i]));
Unitsfi] .CalcLoadsQ;
Unitsfi].UpdateBodyEuter(dt);
} // end i-loop
1
The newline is Units[i].bCollision = CheckForCollisions(&(Units[i]));.Check
ForCollisions is a new function that takes the given unit, whose pointer is passed as
an argument, and checks to see if it’s colliding with anything—in this case, the ground.
If a collision is detected, CheckForCollisions also computes the impact force and re¬
turns true to let us know a collision has occurred. CheckForCollisions is as follows:
boot CheckForColltsions(Particle* p)
1
Vector
n;
Vector
vr;
ftoat
vrn;
ftoat
3;
Vector
Ft;
bool
hasColltston = false;
178 | Chapter 8: Particles
www.it-ebooks.info
// Reset aggregate impact force
p->vImpactForces.x = 0;
p->vImpactForces.y = 0;
// check for collisions with ground plane
if(p->vPosition.y <= (_GROUND_PLANE+p->fRadius)) {
n.x = 0;
n.y = 1;
vr = p->vVelocity;
vrn = vr * n;
// check to see if the particle is moving toward the ground
if(vrn < 0.0) {
3 = -(vr*n) * (-RESTITUTION + 1) * p->fMass;
Fi = n;
Fi *= J/_TIMESTEP;
p->vImpactForces += Fi;
p->vPosition.y = _GROUND_PLANE + p->fRadius;
p->vPosition.x = ((_GROUND_PLANE + p->fRadius -
p->vPreviousPosition.y) /
(p->vPosition.y - p->vPreviousPosition.y) *
(p->vPosition.x - p->vPreviousPosition.x)) +
p->vPreviousPosition.x;
hasCollision = true;
}
}
return hasCollision;
}
CheckForCollisions makes two checks: 1) it checks to see whether or not the particle
is making contact or passing through the ground plane; and 2) it checks to make sure
the particle is actually moving toward the ground plane. Keep in mind a particle could
be in contact with the ground plane right after a collision has been handled with the
particle moving away from the ground. In this case, we don’t want to register another
collision.
Let’s consider the details of this function, starting with the local variables, n is a vector
that represents the unit normal vector pointing from the ground plane to the particle
colliding with it. For ground collisions, in this example, the unit normal vector is always
straight up since the ground plane is flat. This means the unit normal vector will always
have an x component of 0 and its y component will be 1.
The Vector vr is the relative velocity vector between the particle and the ground. Since
the ground isn’t moving, the relative velocity is simply the velocity of the particle, vrn
is a scalar that’s used to store the component of the relative velocity in the direction of
the collision unit normal vector. We compute vrn by taking the dot product of the
relative velocity with the unit normal vector. 3 is a scalar that stores the impulse resulting
from the collision. Fi is a vector that stores the impact force as derived from the impulse
Implementing Collisions | 179
www.it-ebooks.info
3. Finally, hasCollision is a flag that’s set based on whether or not a collision has been
detected.
Now we’ll look at the details within CheckForCollisions. The first task is to initialize
the impact force vector, vlmpactForces, to 0. Next, we make the first collision check by
determining if the y-position of the particle is less than the ground plane height plus
the particles radius. If it is, then we know a collision may have occurred. (_GR0UND_P LAN E
represents they coordinate of the ground plane, which we have set to 100.) If a collision
may have occurred, then we make the next check—to determine if the particle is moving
toward the ground plane.
To perform this second check, we compute the unit normal vector, relative velocity, and
relative velocity component in the collision normal direct as described earlier. If the
relative velocity in the normal direction is negative (i.e., if vrn < 0), then a collision has
occurred. If either of these checks is false, then a collision has not occurred and the
function exits, returning false.
The interesting stuff happens if the second check passes. This is where we have to de¬
termine the impact force that will cause the particle to bounce off the ground plane.
Here’s the specific code that computes the impact force:
J = -(vr*n) * (-RESTITUTION + 1) * p->fMass;
Fi = n;
Fi *= J/_TIMESTEP;
p->vImpactForces += Ft;
p->vPosltion.y = _GROUND_PLANE + p->fRadtus;
p->vPosition.x = (_GROUND_PLANE + p->fRadius -
p->vPreviousPosition.y) /
(p->vPosition.y - p->vPreviousPosition.y) *
(p->vPositton.x - p->vPreviousPosition.x) +
p->vPreviousPosition.x;
hasCollision = true;
We compute the impulse using the formulas presented in Chapter 5. 3 is a scalar equal
to the negative of the relative velocity in the normal direction times the coefficient of
restitution plus 1 times the particle mass. Recall that the coefficient of restitution, _RES
TITUTION, governs how elastic or inelastic the collision is, or in other words, how much
energy is transferred back to the particle during the impact. We have this value set to
0.6, but it is tunable depending on what effect you’re trying to achieve. A value of 1
makes the particles very bouncy, while a value of, say, 0.1 makes them sort of stick to
the ground upon impact.
180 | Chapter 8: Particles
www.it-ebooks.info
Now, to compute the impact force, Fi, that will act on the particle during the next time
step, making it bounce off the ground, we set Fi equal to the collision normal vector.
The magnitude of the impact force is equal to the impulse, J, divided by the time step
in seconds. The line Fi *= J/_TIMESTEP; takes care of calculating the final impact force.
To keep the particle from penetrating the ground, we reposition it so that it’s just resting
on the ground. The y-position is easy to compute as the ground plane elevation plus the
radius of the particle. We compute the x-position by linearly interpolating between the
particles previous position and its current position using the newly computed y-
position. This effectively backs up the particle along the line of action of its velocity to
the point where it is just touching the ground plane.
When you run the simulation now, you’ll see the particles fall, drifting a bit from left to
right until they hit the ground plane. Once they hit, they’ll bounce on the ground,
eventually coming to rest. Their specific behavior in this regard depends on what drag
coefficient you use and what coefficient of restitution you use. If you have wind applied,
when the particles do come to rest, vertically, they should still drift to the right as though
they are sliding on the ground plane.
Partide-to-Obstade Collisions
To make things really interesting, we’ll now add those circular obstacles you saw in
Figure 8-1 through Figure 8-4. The particles will be able to hit them and bounce off or
even settle down into crevasses made by overlapping obstacles. The obstacles are simply
static particles. We’ll define them as particles and initialize them but then skip them
when integrating the equations of motion of the dynamic particles. Here’s the declara¬
tion for the Obstacles array:
Particle Obstacles[_NUM_OBSTACLES];
Initializing the obstacles is a matter of assigning them positions and a common radius
and a mass. The few lines of code shown next were added to the main program’s Initi
alize function to randomly position obstacles in the lower, middle portion of the win¬
dow above the ground plane. Figure 8-1 through Figure 8-4 illustrate how they are
distributed.
bool Initialize(void)
1
for(i=0; 1<_NUM_0BSTACLES; t++)
{
Obstaclesfi].vPosition.x = GetRandomNumber(_WINWIDTH/2 -
_OBSTACLE_RADIUS*10,
_WINWIDTH/2 +
_OBSTACLE_RADIUS*10, false);
Implementing Collisions | 181
www.it-ebooks.info
Obstaclesfi].vPosition.y = GetRandomNurnber(_GROUND_PLANE +
_OBSTACLE_RADIUS, JJINHEIGHT/2 -
_0BSTACLE_RADIUS*4, false);
Obstacles[l].fRadius = _OBSTACLE_RADIUS;
Obstaclesfi].fMass = 100;
}
}
Drawing the obstacles is easy since they are Particle types with a Draw method that
already draws circular shapes. We created DrawObstacles to iterate through the Obsta
cles array, calling the Draw method of each obstacle.
void DrawObstacles(void)
{
int i;
for(i=0; i<_NUM_OBSTACLES; U+)
{
Obstacles[i].Draw();
}
}
DrawObstacles is then called from UpdateSimulatlon:
void UpdateSimulation(void)
{
// initialize the back buffer
if(FrameCounter >= _RENDER_FRAME_COUNT)
{
ClearBackBuffer();
// Draw ground plane
DrawLine(0, _WINHEIGHT - _GROUND_PLANE,
_WINWIDTH, _WINHEIGHT - _GROUND_PLANE,
3, RGB(0,0,0));
DrawObstacles();
}
}
182 | Chapter 8: Particles
www.it-ebooks.info
The last bit of code we need to add to have fully functioning collisions with obstacles
involves adding more collision detection and handling code to the CheckForColli
sions function. Before we look at CheckForCollisions, let’s consider colliding circles
in general to gain a better understanding of what the new code will do.
Figure 8-6 illustrates two circles colliding. We aim to detect whether or not these circles
are colliding by checking the distance between their centers. If the distance between the
two centers is greater than the sum of the radii of the circles, then the particles are not
colliding. The topmost illustration in Figure 8-6 shows the distance, d, between centers
and the distance, s, between the edges of the circles; s is the gap between the two. Another
way to think about this is that if s is positive, then there’s no collision. Referring to the
middle illustration in Figure 8-6, if s is equal to 0, then the circles are in contact. If s is
a negative number, as shown in the bottommost illustration, then the circles are pene¬
trating.
We’ll apply these principles for detecting colliding circles to detecting collisions between
our particles and obstacles since they are both circles. Figure 8-7 illustrates how our
particle-to-obstacle collisions might look.
Implementing Collisions | 183
www.it-ebooks.info
We’ll calculate s for each particle against each obstacle to determine contact or pene¬
tration. If we find either, then we’ll perform the relative velocity check in the same
manner we did for particle-to-ground collisions to see if the particle is moving toward
the obstacle. If it is, then we have a collision and we’ll back up the particle along the
collision normal vector line of action, which is simply the line connecting the centers
of the particle and the obstacle. We’ll also compute the impact force like we did earlier
and let the integrator take care of the rest.
OK, now let’s look at the new code in CheckForCollisions:
bool CheckForColllstons(Particle* p)
{
// Check for collisions with obstacles
float r;
Vector d;
float s;
for(i=0; i<_NUM_OBSTACLES; i++)
{
r = p->fRadius + Obstacles[i].fRadius;
d = p->vPosition - Obstaclesfi].vPosition;
s = d.Magnitude() - r;
184 | Chapter 8: Particles
www.it-ebooks.info
if(s <= 0.0)
{
d.Normalize();
n = d;
vr = p->vVeloctty - Obstaclesfi].vVelocity;
vrn = vr*n;
lf(vrn < 0.0)
{
3 = -(vr*n) * (_RESTITUTION + 1) /
(l/p->fMass + l/Obstacles[i].fMass);
Ft = n;
Ft *= J/JTIMESTEP;
p->vImpactForces += Fi;
p->vPosition -= n*s;
hasCollision = true;
}
}
}
}
The new code is nearly the same as the code that checks for and handles particle-to-
ground collisions. The only major differences are in how we compute the distance be¬
tween the particle and the obstacle and how we adjust the colliding particle’s position
to prevent it from penetrating an obstacle, since the unit normal vector may not be
straight up as it was before. The rest of the code is the same, so let’s focus on the differ¬
ences.
As explained earlier and illustrated in Figure 8-6, we need to compute the separation,
s, between the particle and the obstacle. So to get s, we declare a variable r and equate
it to the sum of radii of the particle and the obstacle against which we’re checking for a
collision. We define d, a Vector, as the difference between the positions of the particle
and obstacle. The magnitude of d minus r yields s.
If s is less than 0, then we make the relative velocity check. Now, in this case the collision
normal vector is along the line connecting the centers of the two circles representing
the particle and the obstacle. Well, that’s just the vector d we already calculated. To get
the unit normal vector, we simply normalize d. The relative velocity vector is simply the
difference in velocities of the particle and the obstacle. Since the obstacles are static, the
relative velocity is just the particle’s velocity. But we calculated the relative velocity by
taking the vector difference vr = p->vVelocity - Obstacles [i] .vVelocity, because
in a more advanced scenario, you might have moving obstacles.
Implementing Collisions | 185
www.it-ebooks.info
Taking the dot product of the relative velocity vector, vr, with the unit normal vector
yields the relative velocity in the collision normal direction. If that relative velocity is
less than 0, then the particle and the object are colliding and the code goes on to calculate
the impact force in a manner similar to that described earlier for the particle-to-ground
collisions. The only difference here is that both the particle’s and object’s masses appear
in the impulse formula. Earlier we assumed the ground was infinitely massive relative
to the particle’s mass, so the 1/m term for the ground went to 0, essentially dropping out
of the equation. Refer back to Chapter 5 to recall the impulse formulas.
Once the impact force is calculated, the code backs up the particle by a distance equal
to s, the penetration, along the line of action of the collision normal vector, giving us
what we desire (as shown in Figure 8-7).
Now, when you run this simulation you’ll see the particles falling down, bouncing off
the obstacles or flowing around them depending on the value you’re using for coefficient
of restitution, ultimately bouncing and coming to rest on the ground plane. If you have
a wind speed greater than 0, then the particles will still drift along the ground plane
from left to right.
Tuning
Tuning is an important part of developing any simulator. Tuning means different things
to different people. For some, tuning is adjusting formulas and coefficients to make your
simulation match some specific “right answer,” while to others tuning is adjusting pa¬
rameters to make the simulation look and behave how you want it to, whether or not
it’s technically the right answer. After all, the right answer for a game is that it’s fun and
robust. Speaking of robustness, other folks view tuning in the sense of adjusting pa¬
rameters to make the simulation stable. In reality, this is all tuning and you should think
of it as a necessary part of developing your simulation. It’s the process by which you
tweak, nudge, and adjust things to make the simulation do what you want it to do.
For example, you can use this same example simulation to model very springy rubber
balls. To achieve this, you’ll probably adjust the coefficient of restitution toward a value
approaching 1 and perhaps lower the drag coefficient. The particles will bounce all over
the place with a lot of energy. If, on the other hand, you want to model something along
the lines of mud, then you’ll lower the coefficient of restitution and increase the drag
coefficient. There is no right or wrong combination of coefficient of restitution or drag
coefficient to use, so long as you are pleased with the results.
Another aspect you might tune is the number of simulation frames per rendering frame.
You may find the simulation calculations take so long that your resulting animations
are too jerky because you aren’t updating the display enough. The converse may be true
in other cases. An important parameter that plays into this is the time step size you take
at each simulation iteration. If the step size is too small, you’ll perform a lot of simulation
186 | Chapter 8: Particles
www.it-ebooks.info
steps, slowing the animation down. On the other hand, a small time step can keep the
simulation numerically stable. Your chosen integrator plays a huge role here.
If you make your time step too large, the simulation may just blow up and not work. It
will be numerically unstable. Even if it doesn’t blow up, you might see weird results. For
example, if the time step in the example simulation discussed in this chapter is too large,
then particles may completely step over obstacles in a single time step, missing the
collision that would otherwise have happened. (We’ll show you in Chapter 10 how to
deal with that situation.)
In general, tuning is a necessary part of developing physics-based simulations, and we
encourage you to experiment—trying different combinations of parameters to see what
results you can achieve. You should even try combinations that may break the example
in this chapter to see what happens and what you should try to avoid in a deployed
game.
Tuning | 187
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 9
2D Rigid-Body Simulator
After reading Chapter 8, you’ve learned the main ingredients that go into a simulator,
specifically a particle simulator. In this chapter we’ll look beyond particles at 2D rigid
bodies. The main difference here is that rigid bodies rotate, and you must deal with an
additional equation of motion—namely, the angular equation of motion relating a rigid
body’s angular acceleration and inertia to the sum of all moments (torques) acting on
the rigid body. The fundamental elements of the simulator—the model, integrator, Ten¬
derer, etc.—are the same as before; you just have to deal with rotation. In two dimen¬
sions, handling rotation is simple. Things get a bit more involved when handling rota¬
tion in three dimensions, and we’ll treat that problem in Chapter 11.
The example we’ll take a close look at in this chapter is simple by design. We want to
focus on the differences between the particle simulator and a 2D rigid-body simulator.
In Chapter 10, we’ll extend this simple example to deal with multiple rigid bodies and
collisions. That’s where things really get interesting. For now, we’ll consider a single rigid
body, a virtual hovercraft, that moves around the screen under the influences of thrust
forces that you can control with the keyboard. While simple, this example covers the
most fundamental aspects of simulating 2D rigid bodies.
Figure 9-1 shows our virtual hovercraft. The pointy end is the front, and the hovercraft
will start off moving from the left side of the screen to the right. Using the arrow keys,
you can increase or decrease its speed and make it turn left or right (port or starboard).
189
www.it-ebooks.info
2
-> x (0°)
+ Rotations
t
y (90°)
Figure 9-1. 2D rigid-body example
In this simulation, the world coordinate system has its x-axis pointing to the right, its
y-axis pointing down toward the bottom of the screen, and the z-axis pointing into the
screen. Even though this is a 2D example where all motion is confined to the x-y plane,
you still need a z-axis about which the hovercraft will rotate. Also, the local, or body-
fixed, coordinate system has its x-axis pointing toward the front of the hovercraft, its
y-axis pointing to the starboard side, and its z-axis into the screen. The local coordinate
system is fixed to the rigid body at its center of gravity location.
Model
The hovercraft modeled in this simulation is a simplified version of the hovercraft we’ll
model in Chapter 17. You can refer to Chapter 17 for more details on that model. For
convenience we repeat some of the basic properties of the model here. Figure 9-2 illus¬
trates the main features of the model.
190 | Chapter 9:2D Rigid-Body Simulator
www.it-ebooks.info
Figure 9-2. Simple hovercraft model
We’re assuming this hovercraft operates over smooth land and is fitted with a single
airscrew propeller, located toward the aft end of the craft, that provides forward thrust.
For controllability, the craft is fitted with two bow thrusters, one to port and the other
to starboard. These bow thrusters are used to steer the hovercraft.
We use a simplified drag model where the only drag component is due to aerodynamic
drag on the entire craft with a constant projected area. This model is similar to the one
used in Chapter 8 for particle drag. A more rigorous model would consider the actual
projected area of the craft as a function of the direction of relative velocity, as in the
flight simulator example discussed in Chapter 15, as well as the frictional drag between
the bottom of the crafts skirt and the ground. We also assume that the center of drag—
the point through which we can assume the drag force vector is applied—is located some
distance aft of the center of gravity so as to give a little directional stability (that is, to
counteract rotation). This serves the same function as the vertical tail fins on aircraft.
Again, a more rigorous model would include the effects of rotation on aerodynamic
drag, but we ignore that here.
In code, the first thing you need to do to represent this vehicle is define a rigid-body
class that contains all of the information you’ll need to track it and calculate the forces
Model | 191
www.it-ebooks.info
and moments acting on it. This RigidBody2D class is very similar to the Particle class
from Chapter 8, but with some additions mostly dealing with rotation. Here’s how we
did it:
class RigidBody2D {
public:
float fMass;
float flnertia;
float
Vector
Vector
Vector
Vector
// total mass (constant)
// mass moment of inertia
flnertialnverse;
vPosition;
vVelocity;
vVelocityBody;
vAngularVelocity;
// inverse of mass moment of inertia
// position in earth coordinates
// velocity in earth coordinates
// velocity in body coordinates
// angular velocity in body coordinates
float
float
fSpeed;
fOrientation;
// speed
// orientation
Vector
Vector
vForces;
vMoment;
// total force on body
// total moment on body
float ThrustForce; // Magnitude of the thrust force
Vector PThrust, SThrust; // bow thruster forces
float
float
float
fWidth;
fLength;
fHeight;
// bounding dimensions
Vector CD; // location of center of drag in body coordinates
Vector CT; // location of center of propeller thrust in body coords.
Vector CPT; // location of port bow thruster thrust in body coords.
Vector CST; // location of starboard bow thruster thrust in body
// coords.
float ProjectedArea;
// projected area of the body
RigidBody2D(void);
void CalcLoads(void);
void UpdateBodyEuler(double dt);
void SetThrusters(bool p, bool s);
void ModulateThrust(bool up);
};
The code comments briefly explain each property, and so far you’ve seen all these prop¬
erties somewhere in this book, so we won’t explain them again here. That said, notice
that several of these properties are the same as those shown in the Particle class from
Chapter 8. These properties include fMass, vPosition, vVelocity, fSpeed, vForces,
and fRadius. All of the other properties are new and required to handle the rotational
motion aspects of rigid bodies.
192 | Chapter 9:2D Rigid-Body Simulator
www.it-ebooks.info
The RigidBody2D constructor is straightforward, as shown next, and simply initializes
all the properties to some arbitrarily tuned values we decided worked well. In Chap¬
ter 17, you’ll see how we model a more realistic hovercraft.
RlgtdBody2D::RtgidBody2D(void)
{
fMass = 100;
flnertta = 500;
flnerttalnverse = l/flnertia;
vPositton.x = 0;
vPosltton.y = 0;
fWidth = 10;
fLength = 20;
fHeight = 5;
fOrtentation = 0;
CD.x = -0.25*fLength;
CD.y = 0.0f;
CD.z = 0.0f;
CT.x = -0.5*fLength;
CT.y = 0.0f;
CT.z = 0.0f;
CPT.x = 0.5*fLength;
CPT.y = -0.5*fWidth;
CPT.z = 0.0f;
CST.x = 0.5*fLength;
CST.y = 0.5*fklidth;
CST.z = 0.0f;
ProjectedArea = (fLength + fWidth)/2 * fHeight; // an approximation
ThrustForce = _THRUSTFORCE;
}
As in the particle simulator of Chapter 8, you’ll notice here that the Vector class is
actually a triple (that is, it has three components—x, y, and z). Since this is a 2D example,
the z components will always be 0, except in the case of the angular velocity vector where
only the z component will be used (since rotation occurs only about the z-axis). The
class that we use in this example is discussed in Appendix A. The reason we didn’t write
a separate 2D vector class, one with only x and y components, is because we’ll extend
this code to 3D later and wanted to get you accustomed to using the 3D vector class.
Besides, it’s pretty easy to create a 2D vector class from the 3D class by simply stripping
out the z component.
As with the particle example of Chapter 8, we need a CalcLoads method for the rigid
body. As before, this method will compute all the loads acting on the rigid body, but
now these loads include both forces and moments that will cause rotation. CalcLoads
looks like this:
Model | 193
www.it-ebooks.info
void RigidBody2D::CalcLoads(void)
{
Vector Fb; // stores the sum of forces
Vector Mb; // stores the sum of moments
Vector Thrust; // thrust vector
// reset forces and moments:
vForces.x = 0.0f;
vForces.y = 0.0f;
vForces.z = 0.0f;
//
always
zero
in
2D
vMoment.x = 0.0f;
//
always
zero
in
2D
vMoment.y = 0.0f;
//
always
zero
in
2D
vMoment.z = 0.0f;
Fb.x = 0.0f;
Fb.y = 0.0f;
Fb.z = 0.0f;
Mb.x = 0.0f;
Mb.y = 0.0f;
Mb.z = 0.0f;
// Define the thrust vector, which acts through the craft's CG
Thrust.x = 1.0f;
Thrust.y = 0.0f;
Thrust.z = 0.0f; // zero in 2D
Thrust *= ThrustForce;
// Calculate forces and moments in body space:
Vector vLocalVelocity;
float fLocalSpeed;
Vector vDragVector;
float tmp;
Vector vResultant;
Vector vtmp;
// Calculate the aerodynamic drag force:
// Calculate local velocity:
// The local velocity includes the velocity due to
// linear motion of the craft,
// plus the velocity at each element
// due to the rotation of the craft.
vtmp = vAngularVelocity A CD; // rotational part
vLocalVelocity = vVelocityBody + vtmp;
// Calculate local air speed
fLocalSpeed = vLocalVelocity .MagnitudeQ;
// Find the direction in which drag will act.
// Drag always acts in line with the relative
194 | Chapter 9:2D Rigid-Body Simulator
www.it-ebooks.info
// velocity but in the opposing direction
if(fLocalSpeed > tol)
{
vLocalVelocity.Normalize();
vDragVector = -vLocalVelocity;
// Determine the resultant force on the element,
tmp = 0.5f * rho * fLocalSpeed*fLocalSpeed
* ProjectedArea;
vResultant = vDragVector * _LINEARDRAGCOEFFICIENT * tmp;
// Keep a running total of these resultant forces
Fb += vResultant;
// Calculate the moment about the CG
// and keep a running total of these moments
vtmp = CD A vResultant;
Mb += vtmp;
}
// Calculate the Port & Starboard bow thruster forces:
// Keep a running total of these resultant forces
Fb += PThrust;
// Calculate the moment about the CG of this element's force
// and keep a running total of these moments (total moment)
vtmp = CPT A PThrust;
Mb += vtmp;
// Keep a running total of these resultant forces (total force)
Fb += SThrust;
// Calculate the moment about the CG of this element's force
// and keep a running total of these moments (total moment)
vtmp = CST A SThrust;
Mb += vtmp;
// Now add the propulsion thrust
Fb += Thrust; // no moment since line of action is through CG
// Convert forces from model space to earth space
vForces = VRotate2D(f0rientation, Fb);
vMoment += Mb;
}
The first thing that CalcLoads does is initialize the force and moment variables that will
contain the total of all forces and moments acting on the craft at any instant in time.
Just as we must aggregate forces, we must also aggregate moments. The forces will be
Model | 195
www.it-ebooks.info
used along with the linear equation of motion to compute the linear displacement of
the rigid body, while the moments will be used with the angular equation of motion to
compute the orientation of the body.
The function then goes on to define a vector representing the propeller thrust, Thrust.
The propeller thrust vector acts in the positive (local) x-direction and has a magnitude
defined by ThrustForce, which the user sets via the keyboard interface (we’ll get to that
later). Note that if ThrustForce is negative, then the thrust will actually be a reversing
thrust instead of a forward thrust.
After defining the thrust vector, this function goes on to calculate the aerodynamic drag
acting on the hovercraft. These calculations are very similar to those discussed in
Chapter 17. The first thing to do is determine the relative velocity at the center of drag,
considering both linear and angular motion. You’ll need the magnitude of the relative
velocity vector when calculating the magnitude of the drag force, and you’ll need the
direction of the relative velocity vector to determine the direction of the drag force since
it always opposes the velocity vector. The line vtmp = vAngularVelocity A CD computes
the linear velocity at the drag center by taking the vector cross product of the angular
velocity vector with the position vector of the drag center, CD. The result is stored in a
temporary vector, vtnp, and then added vectorially to the body velocity vector, vVelo
city Body. The result of this vector addition is a velocity vector representing the velocity
of the point defined by CD, including contributions from the body’s linear and angular
motion. We compute the actual drag force, which acts in line with but in a direction
opposing the velocity vector, in a manner similar to that for particles, using a simple
formula relating the drag force to the speed squared, density of air, projected area, and
a drag coefficient. The following code performs this calculation:
vLocalVelocity.Normalize();
vDragVector = -vLocalVelocity;
// Determine the resultant force on the element,
tmp = 0.5f * rho * fLocalSpeed*fLocalSpeed
* ProjectedArea;
vResultant = vDragVector * _LINEARDRAGCOEFFICIENT * tmp;
Note that the drag coefficient, LINEARDRAGCOEFFICIENT, is defined as follows:
#define LINEARDRAGCOEFFICIENT 1.25f
Once the drag force is determined, it gets aggregated in the total force vector as follows:
Fb += vResultant;
In addition to aggregating this force, we must aggregate the moment due to that force
in the total moment vector as follows:
vtmp = CD A vResultant;
Mb += vtmp;
196 | Chapter 9:2D Rigid-Body Simulator
www.it-ebooks.info
The first line computes the moment due to the drag force by taking the vector cross
product of the position vector, to the center of drag, with the drag force vector. The
second line adds this force to the variable, accumulating these moments.
With the drag calculation complete, CalcLoads proceeds to calculate the forces and
moments due to the bow thrusters, which may be active or inactive at any given time.
Fb += PThrust;
vtmp = CPT A PThrust;
Mb += vtmp;
The first line aggregates the port bow thruster force into Fb. PThrust is a force vector
computed in the SetThrusters method in response to your keyboard input. The next
two lines compute and aggregate the moment due to the thruster force. A similar set of
code lines follows, computing the force and moment due to the starboard bow thruster.
Next, the propeller thrust force is added to the running total of forces. Remember, since
the propeller thrust force acts through the center of gravity, there is no moment to worry
about. Thus, all we need is:
Fb += Thrust; // no moment since line of action is through CG
Finally, the total force is transformed from local coordinates to world coordinates via a
vector rotation given the orientation of the hovercraft, and the total forces and moments
are stored so they are available when it comes time to integrate the equations of motion
at each time step.
As you can see, computing loads on a rigid body is a bit more complex than what you
saw earlier when dealing with particles. This, of course, is due to the nature of rigid
bodies being able to rotate. What’s nice, though, is that all this new complexity is en¬
capsulated in CalcLoads, and the rest of the simulator is pretty much the same as when
we’re dealing with particles.
Transforming Coordinates
Let’s talk about transformation from local to world coordinates a bit more since you’ll
see this sort of transform again in a few places. When computing forces acting on the
rigid body, we want those forces in a vector form relative to the coordinates that are
fixed with respect to the hovercraft (e.g., relative to the body’s center of gravity with the
x-axis pointing toward the front ofthe body and the y-axis pointing toward the starboard
side). This simplifies our calculations of forces and moments. However, when integrat¬
ing the equation of motion to see how the body translates in world coordinates, we use
the equations of motion in world coordinates, requiring us to represent the aggregate
force in world coordinates. That’s why we rotated the aggregate force at the end of the
CalcLoads method.
Model | 197
www.it-ebooks.info
In two dimensions, the coordinate transformation involves a little trigonometry as
shown in the following VRotate2D function:
Vector VRotate2D( float angle. Vector u)
{
float x,y;
x = u.x * cos(DegreesToRadians(-angle)) +
u.y * sin(DegreesToRadians(-angle));
y = -u.x * sin(DegreesToRadlans(-angle)) +
u.y * cos(DegreesToRadians(-angle));
return Vector( x, y, 0);
}
The angle here represents the orientation of the local, body fixed coordinate system with
respect to the world coordinate system. When converting from local coordinates to
world coordinates, use a positive angle; use a negative angle when going the other way.
This is just the convention we’ve adopted so transformations from local coordinates to
world coordinates are positive. You can see we actually take the negative of the angle
parameter, so in reality you could do away with that negative, and then transformations
from local coordinates to world coordinates would actually be negative. It’s your pref¬
erence. You’ll see this function used a few more times in different situations before the
end of this chapter.
Integrator
The UpdateBodyEuler method actually integrates the equations of motion for the rigid
body. Since we’re dealing with a rigid body, unlike a particle, we have two equations of
motion: one for translation, and the other for rotation. The following code sample shows
UpdateBodyEuler.
void RigldBody2D: :LlpdateBodyEuler(double dt)
I
Vector a;
Vector dv;
Vector ds;
float aa;
float dav;
float dr;
// Calculate forces and moments:
CalcLoads();
// Integrate linear equation of motion:
a = vForces / fMass;
dv = a * dt;
vVelocity += dv;
198 | Chapter 9:2D Rigid-Body Simulator
www.it-ebooks.info
ds = vVelocity * dt;
vPosition += ds;
// Integrate angular equation of motion:
aa = vMoment.z / flnertia;
dav = aa * dt;
vAngularVeloclty.z += dav;
dr = RadiansToDegrees(vAngularVeloclty.z * dt);
fOrientation += dr;
// Misc. calculations:
fSpeed = vVelocity.MagnitudeQ;
vVelocttyBody = VRotate2D(-fOrientation, vVelocity);
}
As the name of this method implies, we’ve implemented Euler’s method of integration
as described in Chapter 7. Integrating the linear equation of motion for a rigid body
follows exactly the same steps we used for integrating the linear equation of motion for
particles. All that’s required is to divide the aggregate forces acting on a body by the
mass of the body to get the body’s acceleration. The line of code a = vForces /
fMass does just this. Notice here that a is a Vector, as is vForces. fMass is a scalar, and
the / operator defined in the Vector class takes care of dividing each component of the
vForces vector by fMass and setting the corresponding components in a. The change
in velocity, dv, is equal to acceleration times the change in time, dt. The body’s new
velocity is then computed by the line vVelocity += dv. Here again, vVelocity and dv
are Vectors and the += operator takes care of the vector arithmetic. This is the first actual
integration for translation.
The second integration takes place in the next few lines, where we determine the body’s
displacement and new position by integrating its velocity. The line ds = vVelocity *
d t determines the displacement, or change in the body’s position, and the line vPosition
+= ds computes the new position by adding the displacement to the body’s old position.
That’s it for translation.
The next order of business is to integrate the angular equation of motion to find the
body’s new orientation. The line a a = vMoment.z / flnertia; computes the body’s
angular acceleration by dividing the aggregate moment acting on the body by its mass
moment of inertia, aa is a scalar, as is flnertia since this is a 2D problem. In 3D, things
are a bit more complicated, and we’ll get to that in Chapter 11.
We compute the change in angular velocity, dav, a scalar, by multiplying aa by the time
step size, dt. The new angular velocity is simply the old velocity plus the change: vAn
gula rVelocity. z += dav. The change in orientationis equal to the new angular velocity
multiplied by the time step: vAngularVelocity .z * dt. Notice that we convert the
Model | 199
www.it-ebooks.info
change in orientation from radians to degrees here since we’re keeping track of orien¬
tation in degrees. You don’t really have to, so long as you’re consistent.
The last line in UpdateBody Euler computes the body’s linear speed by transforming the
magnitude of its velocity vector to local, body coordinates. Recall in CalcLoads that we
require the body’s velocity in body-fixed coordinates in order to compute the drag force
on the body.
Rendering
In this simple example, rendering the virtual hovercraft is just a little more involved
than rendering the particles in the example from Chapter 8. All we do is draw a few
connected lines using Windows API calls wrapped in our own functions to hide some
of the Windows-specific code. The following code snippet is all we need to render the
hovercraft:
void DrawCraft(RigldBody2D craft, COLORREF clr)
{
Vector vList[5];
doubte wd, Ig;
int t;
Vector vl;
wd = craft.fWtdth;
Ig = craft.fLength;
vList[0].x = tg/2; vLtst[0].y = wd/2;
vList[l].x = -tg/2; vList[l].y = wd/2;
vList[2].x = -lg/2; vList[2].y = -wd/2;
vList[3].x = tg/2; vLtst[3].y = -wd/2;
vList[4].x = tg/2*1.5; vLtst[4].y = 0;
for(i=0; i<5; i++)
1
vl = VRotate2D(craft.f0rlentatfon, vList[i]);
vLtstfi] = vl + craft.vPosttion;
}
DrawLine(vLtst[0].x, vList[0].y, vList[l].x, vLtst[l].y, 2, clr);
DrawLine(vLtst[l].x, vList[l].y, vList[2].x, vLtst[2].y, 2, clr);
DrawLine(vLtst[2].x, vLtst[2].y, vList[3].x, vLlst[3].y, 2, clr);
DrawLine(vList[3].x, vLtst[3].y, vList[4].x, vLlst[4].y, 2, clr);
DrawLine(vLtst[4].x, vList[4].y, vList[0].x, vLtst[0].y, 2, clr);
}
You can use your own rendering code here, of course, and all you really need to pay
close attention to is transforming the coordinates for the outline of the hovercraft from
body to world coordinates. This involves rotating the vertex coordinates from body-
fixed space using the VRotate2D function and then adding the position of the center of
gravity of the hovercraft to each transformed vertex. These lines take care of this coor¬
dinate transformation:
200 | Chapter 9:2D Rigid-Body Simulator
www.it-ebooks.info
for(l=0; i<5; i++)
{
vl = VRotate2D(craft.fOrientation, vLlst[i]);
vLtstfi] = vl + craft.vPosttion;
}
The Basic Simulator
The heart of this simulation is handled by the RigidBody2D class described earlier.
However, we need to show you how that class is used in the context of the main program.
This simulator is very similar to that shown in Chapter 8 for particles, so if you’ve read
that chapter already you can breeze through this section.
First, we define a few global variables as follows:
// Global Variables:
int FraneCounter = 0;
RlgidBody2D Craft;
FraneCounter counts the number of time steps integrated before the graphics display
is updated. How many time steps you allow the simulation to integrate before updating
the display is a matter of tuning. You’ll see how this is used momentarily when we discuss
the UpdateSimulation function. Craft is a RigidBody2D type that will represent our
virtual hovercraft.
For the most part, Craft is initialized in accordance with the RigidBody2D constructor
shown earlier. However, its position is at the origin, so we make a call to the following
Initialize function to locate the Craft in the middle of the screen vertically and on
the left side. We set its orientation to 0 degrees so it points toward the right side of the
screen:
bool Initialize(void)
{
Craft.vPosition.x = JJINWIDTH/10;
Craft.vPosition.y = JJINHEIGHT/2;
Craft, fomentation = 0;
return true;
}
OK, now let’s consider UpdateSimulation as shown in the code snippet below. This
function gets called every cycle through the program’s main message loop and is re¬
sponsible for making appropriate function calls to update the hovercraft’s position and
orientation, as well as rendering the scene. It also checks the states of the keyboard arrow
keys and makes appropriate function calls:
void UpdateSinulatlon(void)
{
double dt = _TIMESTEP;
RECT r;
The Basic Simulator | 201
www.it-ebooks.info
Craft.SetThrusters(false, false);
if (IsKeyDown(VKJJP))
Craft.ModulateThrust(true);
if (IsKeyDown(VK_DOWN))
Craft.ModulateThrust(false);
if (IsKeyDown(VK_RIGHT))
Craft.SetThrusters(true, false);
if (IsKeyDown(VK_LEFT))
Craft.SetThrusters(false, true);
// update the simulation
Craft.UpdateBodyEuler(dt);
if(FrameCounter >= _RENDER_FRAME_COUNT)
{
// update the display
ClearBackBuffer();
DrawCraft(Craft, RGB(0,0,255));
CopyBackBufferToWindow();
FrameCounter = 0;
} else
FrameCounter++;
if(Craft.vPosition.x > _WINWIDTH) Craft.vPosition.x = 0;
if(Craft.vPosition.x < 0) Craft.vPosition.x = _WINWIDTH;
if(Craft.vPosition.y > _WINHEIGHT) Craft.vPosition.y = 0;
if(Craft.vPosition.y < 0) Craft.vPosition.y = _WINHEIGHT;
}
The local variable dt represents the small yet finite amount of time, in seconds, over
which each integration step is taken. The global define _TIMESTEP stores the time step,
which we have set to 0.001 seconds. This value is subject to tuning.
The first action UpdateSimulatton takes is to reset the states of the bow thrusters to
inactive by calling the SetThrusters method as follows:
Craft.SetThrusters(false, false);
Next, the keyboard is polled using the function IsKeyDown. This is a wrapper function
we created to encapsulate the necessary Windows API calls used to check key states. If
the up arrow key is pressed, then the RigidBody2D method ModulateTh rust is called,
as shown here:
Craft.ModulateThrust(true);
202 | Chapter 9:2D Rigid-Body Simulator
www.it-ebooks.info
If the down arrow key is pressed, then ModulateThrust is called, passing false instead
of true.
ModulateThrust looks like this:
void RigidBody2D::ModulateThrust(bool up)
{
double dT = up ? _DTHRUST:-_DTHRUST;
ThrustForce += dT;
if(ThrustForce > _MAXTHRUST) ThrustForce = _MAXTHRUST;
if(ThrustForce < _MINTHRUST) ThrustForce = _MINTHRUST;
}
All it does is increment the propeller thrust force by a small amount, either increasing
it or decreasing it, depending on the value of the up parameter.
Getting back to UpdateSimulation, we make a couple more calls to I sKeyDown, checking
the states of the left and right arrow keys. If the left arrow key is down, then the Rigid
Body2D method SetThrusters is called, passing false as the first parameter and true
as the second parameter. If the right arrow key is down, these parameter values are
reversed. SetThrusters looks like this:
void RigidBody2D::SetThrusters(bool p, bool s)
1
PThrust.x = 0;
PThrust.y = 0;
SThrust.x = 0;
SThrust.y = 0;
if (p)
PThrust.y = _STEERINGFORCE;
if(s)
SThrust.y = -_STEERINGFORCE;
}
It resets the port and starboard bow thruster thrust vectors and then sets them according
to the parameters passed in SetThrusters. If p is true, then a right turn is desired and
a port thrust force, PThrust, is created, pointing toward the starboard side. This seems
opposite of what you’d expect, but it is the port bow thruster that is fired, pushing the
bow of the hovercraft toward the right (starboard) side. Similarly, if s is true, a thrust
force is created that will push the bow of the hovercraft to the left (port) side.
Now with the thrust forces managed, UpdateSimulation makes the call:
Craft.UpdateBodyEuler(dt)
UpdateBodyEuler integrates the equations of motion as discussed earlier.
The next segment of code checks the value of the frame counter. If the frame counter
has reached the defined number of frames (stored in _RENDER_FRAME_COUNT), then the
The Basic Simulator | 203
www.it-ebooks.info
back buffer is cleared to prepare it for drawing upon and ultimately copying to the
screen.
Finally, the last four lines of code wrap the hovercraft’s position around the edges of the
screen.
Tuning
You’ll probably want to tune this example to run well on your computer since we didn’t
implement any profiling for processor speed. Moreover, you should tune the various
parameters governing the behavior of the hovercraft to see how it responds. The way
we have it set up now makes the hovercraft exhibit a soft sort of response to turning—
that is, upon application of turning forces, the craft will tend to keep tracking in its
original heading for a bit even while yawed. It will not respond like a car would turn.
You can change this behavior, of course.
Some things we suggest you play with include the time step size and the various con¬
stants we’ve defined as follows:
#define
_THRUSTFORCE
5.0f
#define
_MAXTHRUST
10.0f
#define
_MINTHRUST
0.0f
#define
_DTHRUST
0.001f
#define
_STEERINGFORCE
3.0f
#define
LINEARDRAGCOEFFICIENT
_THRUSTFORCE is the initial magnitude of the propeller thrust force. _MAXTHRUST and
_MINTHRUST set upper and lower bounds to this force, which is modulated by the user
pressing the up and down arrow keys. _DTHRUST is the incremental change in thrust in
response to the user pressing the up and down arrow keys. _STEERINGFORCE is the mag¬
nitude of the bow thruster forces. You should definitely play with this value to see how
the behavior of the hovercraft changes. Finally, _LINEARDRAGCOEFFICIENT is the drag
coefficient used to compute aerodynamic drag. This is another good value to play with
to see how behavior is affected. Speaking of drag, the location of the center of drag that’s
initialized in the RigidBody2D constructor is a good parameter to change in order to
understand how it affects the behavior of the hovercraft. It influences the craft’s direc¬
tional stability, which affects its turning radius—particularly at higher speeds.
204 | Chapter 9:2D Rigid-Body Simulator
www.it-ebooks.info
CHAPTER 10
Implementing Collision Response
In this chapter, we’ll show you how to add a little excitement to the hovercraft example
discussed in the preceding chapter. Specifically, we’ll add another hovercraft and show
you how to add collision response so that the hovercraft can crash into each other and
bounce off like a couple of bumper cars. This is an important element for many types
of games, so it’s crucial that you understand the code that we’ll present here. Now would
be a good time to go back and review Chapter 5 to refresh your memory on the funda¬
mentals of rigid-body collision response since we’ll use the principles and formulas
discussed there to develop the collision response algorithms for the hovercraft simula¬
tion. In Chapter 8 you saw how to implement linear collision response for particles, and
now we’ll show you how to handle angular effects.
To start simply, we’ll first show you how to implement collision response as if the hov¬
ercraft were a couple of particles just like those in Chapter 8. This approach uses only
linear impulse and does not include angular effects, so the results will be somewhat
unrealistic for these hovercraft; however, this approach is applicable to other types of
problems that you may be interested in (for example, billiard ball collisions). Plus, taking
this approach allows us to show you very clearly the distinction between linear and
angular effects. Including angular effects will make the simulation much more realistic;
when the hovercraft crash into each other, not only will they bounce off each other, but
they will also spin.
Before diving into collisions, let’s add another hovercraft to the example we started in
Chapter 9. Recall in that example, we had a single craft that you could control using the
keyboard. Now, we’ll add another hovercraft that simply moves under constant forward
thrust. Later, when we add collision detection and response you’ll be able to run into
this new hovercraft to alter its course.
Referring back to the example from Chapter 9, we need to add another craft as follows:
RigtdBody2D Craft2;
205
www.it-ebooks.info
We’re calling the new hovercraft, very creatively, Craft2. In the Initialize function,
we must now add the following code:
boot Initialtze(vold)
{
Craft2.vPosttion.x = _WINWIDTH/2;
Craft2.vPosition.y = _WINHEIGHT/2;
Craft2.f0rientation = 90;
}
This new code sample positions the second hovercraft in the middle of the screen and
pointing toward the bottom.
There are a few required changes to UpdateSimulation as well. First, add Craft2.Up
dateBodyEuler(dt); right after the line Craft.UpdateBodyEuler(dt);. Then, add
DrawCraft(Craft2, RGB(200, 200, 0)); after the similar line that draws the first
Craft. Craft2 will be drawn yellow to distinguish it from the first Craft. Finally, add
the following lies at the end of UpdateSimulation:
lf(Craft2.vPosttion.x > J/JINWIDTH) Craft2.vPosition.x = 0;
lf(Craft2.vPosition.x < 0) Craft2.vPosition.x = _WINWIDTH;
if(Craft2.vPosition.y > _WINHEIGHT) Craft2.vPosition.y = 0;
if(Craft2.vPosition.y < 0) Craft2.vPosition.y = _WINHEIGHT;
Now, we can add the code to handle collision detection and response, allowing you to
ram your hovercraft into the new one we just added.
Linear Collision Response
In this section, we’ll show you how to implement simple collision response, assuming
that the two hovercraft are particles. We’re going to implement only bare-minimum
collision detection in this simulation; however, regardless of the level of sophistication
of your collision detection routines, there are very specific pieces of information that
you must collect from your collision detection routine(s) in order for your physics-
based collision response routines to work.
To revise the hovercraft example of the previous chapter to include simple collision
response, you’ll have to modify the UpdateSimulation function and add a couple more
functions: CheckForCollision and Applylmpulse.
Before showing you CheckForCollision, we want to explain what your collision de¬
tection function must do. First, it must let you know whether or not there is a collision
occurring between the hovercraft. Secondly, it must let you know if the hovercraft are
206 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
penetrating each other. Thirdly, if the hovercraft are colliding, it must tell you what the
collision normal vector is and what the relative velocity is between the colliding hov¬
ercraft.
To determine whether or not there is a collision, you need to consider two factors:
• Whether or not the objects are close enough, within numerical tolerances, to be
considered in colliding contact
• What the relative normal velocity is between the objects
If the objects aren’t close to each other, they obviously have not collided. If they are
within your tolerance for contact, then they may be colliding; and if they are touching
and overlapping such that they are moving inside each other, they are penetrating, as
illustrated in Figure 10-1. If your collision detection routine finds that the two objects
are indeed close enough to be in colliding contact, then you have to do another check
on the relative normal velocity to see if they are moving away from each other or toward
each other. A collision occurs when the objects are in contact and the contact points are
moving toward each other.
Penetration is important because if your objects overlap during the simulation, the re¬
sults won’t look realistic—you’ll have one hovercraft moving inside the other. What you
have to do is detect this penetration condition and then back up your simulation, reduce
the time step, and try again. You keep doing this until they are no longer penetrating or
they are within tolerance to be considered colliding.
You need to determine the normal velocity vector of the collision in order to calculate
the collision impulse that will be used to simulate their response to the collision. For
simple cases, determining this normal vector is fairly straightforward. In the case of
particles or spheres, the collision normal is simply along the line that connects the
Linear Collision Response | 207
www.it-ebooks.info
centers of gravity of each colliding object; this is central impact, as discussed in Chap¬
ter 5, and is the same as that used for the particle example in Chapter 8.
Now take a look at the function we’ve prepared for this simulation to check for collisions:
tnt CheckForCollision (pRigidBody2D bodyl, pRigtdBody2D body2)
{
Vector d;
float r;
int retval = 0;
float s;
Vector vl, v2;
float Vrn;
r = bodyl->ColRadius + body2->ColRadlus;
d = bodyl->vPositlon - body2->vPositlon;
s = d.MagnitudeQ - r;
d.Normalize();
vCollisionNormal = d;
vl = bodyl->vVelocity;
v2 = body2->vVelocity;
vRelatlveVelocity = vl - v2;
Vrn = vRelatlveVelocity * vColllsionNornal;
if((fabs(s) <= ctol) && (Vrn < 0.0))
{
retval = 1; // collision;
CollisionBodyl = bodyl;
CollisionBody2 = body2;
} else if(s < -ctol)
{
retval = -1; // interpenetrating
} else
retval = 0; // no collision
return retval;
}
This function uses a simple bounding circle check to determine whether or not the
hovercraft are colliding. The first thing it does is calculate the distance, r, that represents
the absolute minimum separation between these hovercraft when they are in contact.
ColRadius is the radius of the bounding circle of the hovercraft. We must compute it
for each hovercraft upon initialization as follows:
->ColRadius = SQRT(fLength*fLength + fWidth*fWidth);
Next, the distance separating the hovercraft at the time this function is called is deter¬
mined and stored in the variable d. Since we’re assuming that these hovercraft are par¬
ticles, determining d is simply a matter of calculating the distance between the coordi-
208 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
nates of each craft’s center of gravity. In terms of vectors, this is simply the position
vector of one craft minus the position vector of the other.
Once the function has d and r, it needs to determine the actual amount of space, s,
separating the hovercraft’s bounding circles. After this separation is determined, the
function normalizes the vector d. Since the vector d is along the line that separates the
hovercraft’s centers of gravity, normalizing it yields the collision normal vector that we
need for our collision response calculations. The collision normal vector is saved in the
global variable vCollisionNormal.
After calculating the collision normal, this function goes on to determine the relative
velocity between the hovercraft. In vector form, this is simply the difference between
the velocity vectors of each craft. Note that the velocity vectors used here must be in
global coordinates, not body-fixed (local) coordinates. Since what’s really needed to
determine if a collision is made is the relative normal velocity, the function proceeds to
take the vector dot product of the relative velocity and the collision normal vectors,
saving the result in the variable Vrn.
At this point, all of the calculations are complete, and the only thing left to do is make
the appropriate checks to determine if there is a collision, penetration, or no collision
at all.
The first check is to see if the hovercraft are colliding. We determine this by comparing
the absolute value of the separation between the hovercraft, s, with a distance tolerance,
ctol. If the absolute value of s is less than ctol, a collision might be occurring. The
second requirement is that the relative normal velocity be negative, which implies that
the points of impact on the hovercraft are moving toward each other. If there is a col¬
lision, the function returns a 1 to indicate that collision response is necessary.
If the hovercraft are found not to be colliding, then we perform a second check to see
if they’ve moved so close together that they are penetrating each other. In this case, if s
is less than -ctol, the hovercraft are penetrating and the function returns a -1. If the
hovercraft are not colliding and not penetrating, then the function simply returns a 0,
indicating that no further action is required.
Before moving on, let’s say a word or two about ctol—the collision tolerance distance.
This value is subject to tuning. There’s no single value that works well in all cases. You
must consider the overall sizes of the objects potentially colliding, the step size you’re
using, and how far the colliding objects are from the viewer while being rendered (i.e.,
their scale). Basically, you should choose a value that makes collisions look correct, so
that on the one hand objects do not appear to be penetrating each other, and on the
other hand you do not report a collision when objects do not appear to be touching at
all.
Linear Collision Response | 209
www.it-ebooks.info
Take a look now at the other new function. Apply Impulse:
void ApplyImpulse(pRigidBody2D bodyl, pRigidBody2D body2)
{
float j;
j = (-(1+fCr) * (vRelativeVelocity*vCollisionNormal)) /
( (vCollisionNormal*vCollisionNomal) *
(l/bodyl->fMass + l/body2->fMass) );
bodyl->vVeloclty += (j * vColllsionNomal) / bodyl->fMass;
body2->vVeloclty -= (j * vColllsionNomal) / body2->fMass;
}
This is a simple but crucial function for collision response. What it does is calculate the
linear collision impulse as a function of the colliding hovercraft’s relative normal ve¬
locity, masses, and coefficient of restitution, using the formula that we showed you in
Chapter 5. Further, it applies this impulse to each hovercraft, effectively changing their
velocities in response to the collision. Note that the impulse is applied to one hovercraft
and then the negative impulse applied to the other.
With those two new functions complete, it’s now time to revise UpdateSimulation to
handle collision detection and response as the simulation steps through time. Here’s
what the new UpdateSimulation function looks like:
void UpdateSimulation(float dt)
{
float dtime = dt;
bool tryAgain = true;
int check=0;
RigldBody2D craftlCopy, craft2Copy;
bool didPen = false;
int count = 0;
Craft.SetThrusters(false, false);
if (IsKeyDown(VK_UP))
Craft,ModulateThrust(true);
if (IsKeyDown(VK_DOWN))
Craft,ModulateThrust(false);
if (IsKeyDown(VK_RIGHT))
Craft.SetThrusters(true, false);
if (IsKeyDown(VK_lEFT))
Craft.SetThrusters(false, true);
while(tryAgain && dtime > tol)
210 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
{
tryAgain = false;
memcpy(&craftlCopy, &Craft, sizeof(RtgidBody2D));
memcpy(&craft2Copy, &Craft2, sizeof(RigidBody2D));
Craft.UpdateBodyEuler(dtlne);
Craft2.UpdateBodyEuler(dtime);
CollisionBodyl = 0;
ColllsionBody2 = 0;
check = CheckForCollision(&craftlCopy, &craft2Copy);
If(check == PENETRATING)
{
dtime = dtime/2;
tryAgain = true;
didPen = true;
} else tf(check == COLLISION)
{
if(CollisionBodyl != 0 && CollisionBody2 != 0)
ApplyImpulse(CollisionBodyl, CollisionBody2);
}
}
if(IdidPen)
{
memcpy(&Craft, &craftlCopy, sizeof(RigidBody2D));
memcpy(&Craft2, &craft2Copy, sizeof(RigidBody2D));
}
}
Obviously, this version is more complicated than the original version. There’s one main
reason for this: penetration could occur because the hovercraft can move far enough
within a single time step to become overlapped. Visually, this situation is unappealing
and unrealistic, so you should to try to prevent it.
The first thing this function does is enter a while loop:
while(tryAgain && dtime > tol)
{
}
This loop is used to back up the simulation if penetration has occurred on the initial
time step. What happens is this: the function first tries to update the hovercraft and then
checks to see if there is a collision. If there is a collision, then it gets handled by applying
the impulse. If there is penetration, however, then you know the time step was too big
and you have to try again. When this occurs, tryAgain is set to true, the time step is
cut in half, and another attempt is made. The function stays in this loop as long as there
Linear Collision Response | 211
www.it-ebooks.info
is penetration or until the time step has been reduced to a size small enough to force an
exit to the loop. The purpose of this looping is to find the largest step size, less than or
equal to dt, that can be taken and still avoid penetration. You either want a collision or
no collision.
You might ask yourself when does small become too small in terms of time step? Too
small is obviously when the time step approaches 0 and your entire simulation grinds
to a halt. Therefore, you may want to put in some criteria to exit this loop before things
slow down too much. This is all subject to tuning, by the way, and it also depends on
the value you set for ctol. We can’t stress enough the importance of tuning these pa¬
rameters. Basically, you must strive for visual realism while keeping your frame rates
up to required levels.
Looking inside this while loop reveals what’s going on. First, tryAgain is set to false,
optimistically assuming that there will be no penetration, and we make copies of the
hovercraft’s states, reflecting the last successful call to UpdateSimulation.
Next, we make the usual call to UpdateBody for each copy of the hovercraft. Then a call
to the collision detection function, CheckForCollision, is made to see if Craft is col¬
liding with or penetrating Craft2. If there is penetration, then tryAgain is set to true,
dtime is cut in half, didPen is set to true, and the function takes another lap through
the while loop. didPen is a flag that lets us know that a penetration condition did occur.
If there was a collision, the function handles it by applying the appropriate impulse:
if(ColtistonBodyl != 0 && CollisionBody2 != 0)
ApplyInputse(CollistonBodyl, CollisionBody2);
After getting through the while loop, the updated hovercraft states are saved and Upda
teSimulation is complete.
The last bit of code you need to add includes a few new global variables and defines:
#define LINEARDRAGCOEFFICIENT 0.25f
#define COEFFICIENTOFRESTITUTION 0.5f
#define COLLISIONTOLERANCE 2.0f
Vector vColllsionNormal;
Vector vRelatlveVelocity;
float fCr = COEFFICIENTOFRESTITUTION;
float const ctol = COLLISIONTOLERANCE;
The only one we haven’t mentioned so far, although you’ve seen it in Applylmpulse, is
fCr, the coefficient of restitution. Here we have it set to 0.5, which means that the
collisions are halfway between perfectly elastic and perfectly inelastic (refer back to our
earlier discussions on coefficients of restitution in Chapter 5 if you’ve forgotten these
terms). This is one of those parameters that you’ll have to tune to get the desired be¬
havior.
212 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
While we’re on the subject of tuning, we should mention that you’ll also have to play
with the linear drag coefficient used to calculate the drag force on the hovercraft. While
this coefficient is used to simulate fluid dynamic drag, it also plays an important role in
terms of numerical stability. You need some damping in your simulation so that your
integrator does not blow up—that is, damping helps keep your simulation stable.
That’s pretty much it as far as implementing basic collision response. If you run this
example, you’ll be able to drive the hovercraft into each other and bounce off accord¬
ingly. You can play around with the mass of each hovercraft and the coefficient of res¬
titution to see how the craft behave when one is more massive than the other, or when
the collision is somewhere between perfectly elastic and perfectly inelastic.
You may notice that the collision response in this example sometimes looks a little
strange. Keep in mind that’s because this collision response algorithm, so far, assumes
that the hovercraft are round when in fact they are rectangular. This approach will work
just fine for round objects like billiard balls, but to get the level of realism required for
non-round rigid bodies you need to include angular effects. We’ll show you how to do
that in the next section.
Angular Effects
Including angular effects will yield more realistic collision responses for these rigid
bodies, the hovercraft. To get this to work, you’ll have to make several changes to
Applylmpulse and CheckForCollision;. UpdateSimulation will remain unchanged.
The more extensive changes are in CheckForCollision, so we’ll discuss it first.
The new version of CheckForCollision will do more than a simple bounding circle
check. Here, each hovercraft will be represented by a polygon with four edges and four
vertices, and the types of contact that will be checked for are vertex-vertex and vertex-
edge contact (see Figure 10-2). 1
1. Note that this function does not handle multiple contact points.
Angular Effects | 213
www.it-ebooks.info
In addition to the tasks discussed in the last section, this new version of CheckForCol
lision must also determine the exact point of contact between the hovercraft. This is
a very important distinction between this new version and the last. You need to know
the point of contact because in order to affect the angular velocity, you must apply the
impulse at the point of contact. In the last section, the normal to the contact point always
passed through the center of gravity of the hovercraft because we assumed they were
spheres; that’s not the case here.
This now brings up the challenge of finding the collision normal. There are two cases
to consider here. In edge-vertex collisions, the normal is always perpendicular to the
edge that’s involved in the collision. In vertex-vertex collisions, however, the normal is
ambiguous, so we’ve resorted to taking the normal parallel to the line connecting the
hovercraft’s centers of gravity.
All of these considerations make CheckForCollisions a little more involved than in
the previous section. The following code listing shows what we mean:
214 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
int CheckForCollision(pRigidBody2D bodyl, pRigidBody2D body2)
{
Vector
d;
float
r;
int
retval = 0;
float
s;
Vector
vListl[4], vList2[4];
float
wd, Ig;
int
t, j;
bool
haveNodeNode = false;
bool
interpenetrating = false;
bool
haveNodeEdge = false;
Vector
vl, v2, u;
Vector
edge, p, proj;
float
dist, dot;
float
Vrn;
// First check to see if the bounding circles are colliding
r = bodyl->fLength/2 + body2->fLength/2;
d = bodyl->vPosition - body2->vPosition;
s = d.MagnitudeQ - r;
if(s
{
<= ctol)
// We have a possible collision, check further
// build vertex lists for each hovercraft
wd = bodyl->fWidth;
Ig = bodyl->fLength;
vListl[0].y = wd/2; vListl[0].x = lg/2;
vListl[l].y = -wd/2; vListl[l].x = lg/2;
vListl[2].y = -wd/2; vListl[2].x = -lg/2;
vListl[3].y = wd/2; vListl[3].x = -lg/2;
for(l=0; i<4; i++)
{
VRotate2D(bodyl->f0rientation, vListl[i]);
vListlfi] = vListlfi] + bodyl->vPosition;
}
wd = body2->fWidth;
Ig = body2->fLength;
vList2[0].y = wd/2;
vList2[l].y = -wd/2;
vList2[2].y = -wd/2;
vList2[3].y = wd/2;
vList2[0].x = lg/2;
vList2[l].x = lg/2;
vList2[2].x = -lg/2;
vLlst2[3].x = -lg/2;
for(t=0; i<4; i++)
{
VRotate2D(body2->fOrientation, vList2[i]);
vList2[i] = vList2[i] + body2->vPosition;
}
// Check for vertex-vertex collision
Angular Effects | 215
www.it-ebooks.info
for(i=0; i<4 && !haveNodeNode; i++)
{
for(j=0; j<4 && !haveNodeNode; j++)
{
vColllsionPoint = vListl[l];
bodyl->vCollisionPoint = vColllsionPoint -
bodyl->vPosition;
body2->vCollisionPoint = vColllsionPoint -
body2->vPosition;
vCollisionNornal = bodyl->vPosition -
body2->vPosition;
vCollisionNornal.Nornalize();
vl = bodyl->vVelocityBody +
(body 1 - >vAngula rVelocity A body 1 - >vCollisionPoint);
v2 = body2->vVelocityBody +
(body 2 - >vAngula rVelocity A body 2 - >vCollisionPoin t);
vl = VRotate2D(bodyl->f0rientation, vl);
v2 = VRotate2D(body2->fOrientation, v2);
vRelativeVelocity = vl - v2;
Vrn = vRelativeVelocity * vCollisionNornal;
if( ArePointsEqual(vListl[i],
vList2[j]) &&
(Vrn < 0.0) )
haveNodeNode = true;
}
}
// Check for vertex-edge collision
if(!haveNodeNode)
{
for(i=0; i<4 && !haveNodeEdge; i++)
{
for(j=0; j<3 && !haveNodeEdge; j++)
{
if(j--2)
edge = vList2[0] - vList2[j];
else
edge = vList2[j+l] - vList2[j];
u = edge;
u.Nornalize();
p = vListl[i] - vList2[j];
216 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
proj = (p * u) * u;
d = p A u;
dist = d.Magnltude();
vCollislonPoint = vl_istl[i];
bodyl->vColllstonPotnt = vColllstonPolnt -
body1->vPosition;
body2->vColllstonPolnt = vCollisionPoint -
body2->vPosltlon;
vCollislonNornal = ((u A p) A u);
vCollislonNornal.Normallze();
vl = bodyl->vVelocityBody +
(bodyl->vAngularVelocity A
bodyl->vColllsionPolnt);
v2 = body2->vVelocltyBody +
(body2->vAngularVelocity A
body2->vColllslonPolnt);
vl = VRotate2D(bodyl->f0rlentatlon, vl);
v2 = VRotate2D(body2->f0rlentatlon, v2);
vRelati.veVeloci.ty = (vl - v2);
Vrn = vRelativeVelocity * vColltstonNornal;
if( (proj.Magnitude() > 0.0f) &&
(proj.Magnitude() <= edge.Magnitude()) &&
(dtst <= ctol) &&
(Vrn < 0.0) )
haveNodeEdge = true;
}
}
}
// Check for penetration
If(!haveNodeNode && ! haveNodeEdge)
{
for(l=0; t<4 && Unterpenetratlng; 1++)
{
for(j=0; j<4 && Unterpenetratlng; j++)
{
tf(j==3)
edge = vLlst2[0] - vLlst2[j];
else
edge = vLlst2[j+l] - vLlst2[j];
p = vLlstl[l] - vLlst2[j];
dot = p * edge;
Angular Effects | 217
www.it-ebooks.info
lf(dot < 0)
{
If(Interpenetrating)
{
interpenetrating = true;
retval = -1;
} eise If(haveNodeNode || haveNodeEdge)
{
retval = 1;
} else
retval = 0;
} else
{
retval = 0;
}
return retval;
}
The first thing that CheckForCollision does is perform a quick bounding-circle check
to see if there is a possible collision. If no collision is detected, the function simply exits,
returning 0. This is the same bounding-circle check performed in the earlier version:
r = bodyl->fLength/2 + body2->fLength/2;
d = bodyl->vPosition - body2->vPosition;
s = d.MagnitudeQ - r;
if(s <= ctol)
{
} else
retval = 0;
}
If the bounding-circle check indicates the possibility of a collision, then CheckForCol
lision proceeds by setting up a couple of polygons, represented by vertex lists, for each
hovercraft:
wd = bodyl->fWidth;
lg = bodyl->fLength;
vLtstl[0].y = wd/2;
vListlfl].y = -wd/2;
vListl[2].y = -wd/2;
vListl[3].y = wd/2;
vLlstl[0].x = lg/2;
vLlstl[l].x = lg/2;
vListl[2].x = -lg/2;
vListl[3].x = -lg/2;
218 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
for(i=0; i<4; i++)
{
VRotate2D(bodyl->fOrientation, vListl[i]);
vListl[i] = vListlfi] + bodyl->vPosition;
}
wd = body2->fWidth;
Ig = body2->fLength;
vList2[0].y = wd/2;
vList2[l].y = -wd/2;
vList2[2].y = -wd/2;
vList2[3].y = wd/2;
for(i=0; i<4; i++)
{
vList2[0].x = lg/2;
vList2[l].x = lg/2;
vList2[2] .x = -lg/2;
vLlst2[3].x = -lg/2;
VRotate2D(body2->fOrientation, vList2[i]);
vList2[i] = vList2[i] + body2->vPosition;
}
The vertex lists are initialized in unrotated body-fixed (local) coordinates based on the
length and width of the hovercraft. The vertices are then rotated to reflect the orientation
of each hovercraft. After that, the position of each hovercraft is added to each vertex to
convert from local coordinates to global coordinates
Checking first for vertex-vertex collisions, the function iterates through each vertex in
one list, comparing it with each vertex in the other list to see if the points are coincident.
// Check for vertex-vertex collision
for(i=0; i<4 && !haveNodeNode; i++)
{
for(j=0; j<4 && !haveNodeNode; j++)
{
vCollisionPoint = vListlfi];
bodyl->vCollisionPoint = vCollisionPoint -
bodyl->vPosition;
body2->vCollisionPoint = vCollisionPoint -
body2->vPosition;
vCollisionNornal = bodyl->vPosition -
body2->vPosition;
vCollisionNornal.Nornalize();
vl = bodyl->vVelocityBody +
(bodyl->vAngularVelocity A bodyl->vCollisionPoint);
v2 = body2->vVelocityBody +
(body 2 - >vAngula rVelocity A body 2 - >vCollisionPoin t);
vl = VRotate2D(bodyl->f0rientation, vl);
v2 = VRotate2D(body2->fOrientation, v2);
Angular Effects | 219
www.it-ebooks.info
vRelativeVelocity = vl - v2;
Vrn = vRelativeVelocity * vCollisionNormal;
if( ArePointsEqual(vListl[i],
vList2[j]) &&
(Vrn < 0.0) )
haveNodeNode = true;
}
}
This comparison makes a call to another new function, ArePointsEqual:
tf( ArePointsEqual(vListl[i],
vList2[j]) &&
(Vrn < 0.0) )
haveNodeNode = true;
ArePointsEqual simply checks to see if the points are within a specified distance from
each other, as shown here:
bool ArePointsEqual(Vector pi, Vector p2)
{
// Points are equal if each component is within ctol of each other
if( (fabs(pl.x - p2.x) <= ctol) &&
(fabs(pl.y - p2.y) <= ctol) &&
(fabs(pl.z - p2.z) <= ctol) )
return true;
else
return false;
}
Within the nested for loops of the vertex-vertex check, we perform a number of im¬
portant calculations to determine the collision normal vector and relative velocity that
are required for collision response.
First, we calculate the collision point, which is simply the coordinates of a vertex that is
involved in the collision. Note that this point will be in global coordinates, so it will have
to be converted to local coordinates for each hovercraft in order to be useful for collision
response. Here’s how that’s done:
vCollislonPoint = vListlfi];
bodyl->vCollislonPoint = vCollisionPoint -
bodyl->vPosition;
body2->vCollislonPoint = vCollisionPoint -
body2->vPosition;
The second calculation is aimed at determining the collision normal vector, which for
vertex-vertex collisions we’ve assumed is along the line connecting the centers of gravity
220 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
of each hovercraft. The calculation is the same as that shown in the earlier version of
CheckForCollision:
vCollisionNormal = bodyl->vPosition -
body2->vPosition;
vCollistonNormal.Normalize();
The third and final calculation is aimed at determining the relative velocity between the
points of impact. This is an important distinction from the earlier version, since the
velocities of the points of impact on each body are functions of the linear and angular
velocities of the hovercraft:
vl = bodyl->vVelocityBody +
(bodyl->vAngularVelocity A bodyl->vCotlisionPotnt);
v2 = body2->vVelocityBody +
(body2->vAngularVelocity A body2->vCollisionPoint);
vl = VRotate2D(bodyl->f0rientatlon, vl);
v2 = VRotate2D(body2->fOrientatlon, v2);
vRelativeVetocity = vl - v2;
Vrn = vRelativeVetocity * vCollisionNomal;
Here, vl and v2 represent the velocities of the points of collision relative to each hov¬
ercraft in local coordinates, which are then converted to global coordinates. Once we’ve
obtained the relative velocity, vRelativeVelocity, we obtain the relative normal ve¬
locity, Vrn, by taking the dot product of the relative velocity with the collision normal
vector.
If there is no vertex-vertex collision, CheckForCollision proceeds to check for vertex-
edge collisions:
// Check for vertex-edge collision
if(!haveNodeNode)
{
for(i=0; i<4 && !haveNodeEdge; i++)
{
for(j=0; j<3 && !haveNodeEdge; j++)
{
if(j==3)
edge = vList2[0] - vList2[j];
else
edge = vList2[j+l] - vList2[j];
u = edge;
u.NormalizeO;
p = vListl[i] - vList2[j];
proj = (p * u) * u;
d = p A u;
Angular Effects | 221
www.it-ebooks.info
dlst = d.Magnitude();
vCollisionPoint = vLlstl[i];
bodyl->vCollisionPoint = vCollisionPoint -
bodyl->vPosition;
body2->vCollisionPoint = vCollisionPoint -
body2->vPosition;
vCollisionNornal = ((u A p) A u);
vCollisionNornal. Nornalize();
vl = bodyl->vVelocityBody +
(bodyl->vAngularVelocity A
bodyl->vCollisionPoint);
v2 = body2->vVelocityBody +
(body2->vAngularVelocity A
body2->vCollisionPoint);
vl = VRotate2D(bodyl->f0rientation, vl);
v2 = VRotate2D(body2->f0rientation, v2);
vRelativeVelocity = (vl - v2);
Vrn = vRelativeVelocity * vCollisionNornal;
if( (proj.Magnitude() > 0.0f) &&
(proj.Magnitude() <= edge.MagnitudeQ) &&
(dist <= ctol) &&
(Vrn < 0.0) )
haveNodeEdge = true;
}
}
}
Here, the nested f o r loops check each vertex in one list to see if it is in contact with each
edge built from the vertices in the other list. After building the edge under consideration,
we save and normalize a copy of it to represent a unit vector pointing along the edge:
if(j==3)
edge = vList2[0] - vList2[j];
else
edge = vList2[j+l] - vList2[j];
u = edge;
u.Nornalize();
Variable u represents that unit vector, and it will be used in subsequent calculations. The
next set of calculations determines the location of the projection of the vertex under
consideration onto the edge under consideration, as well as the minimum distance from
the vertex to edge:
222 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
p = vListl[i] - vList2[j];
proj = (p * u) * u;
d = p A u;
dist = d.Magni.tude();
Variable p is a vector from the first vertex on the edge to the vertex under consideration,
and proj is the distance from the first edge vertex, along the edge, to the point upon
which the vertex projects, dist is the minimum distance from the vertex to the edge.
Figure 10-3 illustrates this geometry.
If there is a collision, the global location of the point of impact is equal to the vertex
under consideration, which we must convert to local coordinates for each hovercraft,
as shown here:
vCollistonPoint = vListl[i];
bodyl->vCollisionPotnt = vColltsionPolnt -
bodyl->vPosition;
body2->vCollisionPotnt = vCollistonPoint -
body2->vPosition;
Since, in this type of collision, the collision normal vector is perpendicular to the edge,
you can determine it by taking the result of the cross product of u and p and crossing it
with u as follows:
vCollisionNormal = ((u A p) A u);
vCollisionNormal. Nomaltze();
These calculations give you a unit length vector in the plane of vectors u and p and
perpendicular to the edge.
Next, the relative velocity between the points of impact on each hovercraft is determined,
just as in the vertex-vertex collision check:
vl = bodyl->vVelocityBody +
(bodyl->vAngularVelocity A
bodyl->vColtisionPoint);
Angular Effects | 223
www.it-ebooks.info
v2 = body2->vVelocttyBody +
(body2->vAngularVelocity
body2->vCollisionPoint);
vl = VRotate2D(bodyl->f0rientatlon, vl);
v2 = VRotate2D(body2->f0rientatlon, v2);
vRelatlveVelocity = (vl - v2);
Vrn = vRelatlveVelocity * vCollistonNormal;
In determining whether or not the vertex under consideration is in fact colliding with
an edge, you have to check to see if the distance from the vertex is within your collision
tolerance, and you also have to make sure the vertex actually projects onto the edge (that
is, it does not project beyond the endpoints of the edge). Additionally, you need to make
sure the relative normal velocity indicates that the points of contact are moving toward
each other. Here’s how this check looks:
if( (proj.Magnitude() > 0.0f) &&
(proj.Magnltude() <= edge.MagnitudeO) &&
(dlst <= ctol) &&
(Vrn < 0.0) )
haveNodeEdge = true;
After CheckForCollision checks for vertex-vertex and vertex-edge collisions, it goes
on to check for penetration:
if(!haveNodeNode && !haveNodeEdge)
1
for(l=0; !<4 && Unterpenetrating; 1++)
{
for(j=0; j<4 && linterpenetrating; j++)
{
tf(j==3)
edge = vList2[0] - vList2[j];
else
edge = vList2[j+l] - vList2[j];
p = vListlfi] - vList2[j];
dot = p * edge;
if(dot < 0)
{
interpenetrating = true;
}
}
}
1
This check is a standard point-in-polygon check using the vector dot product to deter¬
mine if any vertex of one polygon lies within the bounds of the other polygon. After
this check, the function simply returns the appropriate result. Here again, 0 indicates
no collision or penetration, 1 indicates a collision, and -1 indicates penetration.
224 | Chapter 10: Implementing Collision Response
www.it-ebooks.info
With CheckForCollision out of the way, turn your attention to Applylmpulse, which
also has to be revised to include angular effects. Specifically, you need to use the impulse
formula that includes angular as well as linear effects (see Chapter 5), and you also have
to apply the impulse to the hovercraft’s angular velocities in addition to their linear
velocities. Here’s how the new Applylmpulse function looks:
void ApplyImpulse(pRlgidBody2D bodyl, pRigidBody2D body2)
{
float j;
j = (-(1+fCr) * (vRelatlveVeloci.ty*vColllsionNormal)) /
( (l/bodyl->fMass + l/body2->fMass) +
(vCollisionNormal * (((bodyl->vCollisionPoint A
vCollisionNormal)/bodyl->fInertia) A bodyl->vCollisionPoint)) +
(vCollisionNormal * (((body2->vCollisionPoint A
vCollisionNormal)/body2->fInertia) A body2->vCollisionPoint))
);
bodyl->vVelocity += (j * vCollisionNormal) / bodyl->fMass;
bodyl->vAngularVelocity += (bodyl->vCollisionPoint A
(j * vCollisionNormal)) /
bodyl->fInertia;
body2->vVelocity -= (j * vCollisionNormal) / body2->fMass;
body2->vAngularVelocity -= (body2->vCollisionPoint A
(j * vCollisionNormal)) /
body2->fInertia;
}
Remember, the impulse is applied to one hovercraft while its negative is applied to the
other.
That does it for this new version of the hovercraft simulation. If you run the program
now, you’ll see that you can crash the hovercraft into each other and they bounce and
rotate accordingly. This makes for a much more realistic simulation than the simple,
linear collision response approach of the last section. Here again, you can play with the
mass of each hovercraft and the coefficient of restitution to see how these parameters
affect the collision response between the hovercraft.
Angular Effects | 225
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 11
Rotation in 3D Rigid-Body Simulators
A fundamental difference between particles and rigid bodies is that we cannot ignore
rotation of rigid bodies. This applies to both 2D and 3D rigid bodies. In two dimensions,
it’s quite easy to express the orientation of a rigid body; you need only a single scalar to
represent the body’s rotation about a single axis. In three dimensions, however, there
are three primary coordinate axes about each of which a rigid body may rotate. More¬
over, a rigid body in three dimensions may rotate about any arbitrary axis, not neces¬
sarily one of the coordinate axes.
In two dimensions, we say that a rigid body has only one rotational degree of freedom,
whereas in three dimensions we say that a rigid body has three rotational degrees of
freedom. This may lead you to infer that in three dimensions, you must have three scalar
quantities to represent a body’s rotation. Indeed, this is a minimum requirement, and
you’re probably already familiar with a set of angles that represent the orientation of a
rigid body in 3D—namely, the three Euler angles (roll, pitch, and yaw) that we’ll talk
about in Chapter 15.
These three angles—roll, pitch, and yaw—are very intuitive and easy for us to visualize.
For example, in an airplane the nose pitches up or down, the plane rolls (or banks) left
or right, and the yaw (or heading) changes to the left or right. Unfortunately, there’s a
problem with using these three Euler angles in rigid-body simulations. The problem is
a numerical one that occurs when the pitch angle reaches plus or minus 90 degrees (tt/
2). When this happens, roll and yaw become ambiguous. Worse yet, the angular equa¬
tions of motion written in terms of Euler angles contain terms involving the cosine of
the pitch angle in the denominator, which means that when the pitch angle is plus or
minus 90 degrees the equations become singular (i.e., there’s division by 0). If this hap¬
pens in your simulation, the results would be unpredictable to say the least. Given this
problem with Euler angles, you must use some other means of keeping track of orien¬
tation in your simulation. We’ll discuss two such means in this chapter—specifically,
rotation matrices and quaternions.
227
www.it-ebooks.info
Virtually every computer graphics book that we’ve read contains a chapter or section
on using rotation matrices. Far fewer discuss quaternions, but if you’re familiar with
quaternions, it’s probably in the same context as rotation matrices—that is, how they
are used to rotate 3D points, objects, scenes, and points of view. In a simulation, however,
you need to get a little more out of rotation matrices or quaternions and will use them
in a different context than what you might be accustomed to. Specifically, you need to
keep track of a body’s orientation in space and, moreover, the change in orientation over
time. So it’s in this light that we’ll discuss rotation matrices and quaternions. We’ll try
to be as concise as possible so as not to cloud the water with the proofs and derivations
that you can find in the texts referred to in the Bibliography.
Rotation Matrices
A rotation matrix is a 3x3 matrix that, when multiplied with a point or vector, results
in the rotation of that point about some axis, yielding a new set of coordinates. You can
rotate points about axes in one coordinate system or you can use rotation matrices to
convert points from one coordinate system to another, where one is rotated relative to
the other.
Rotating a vector by a rotation matrix is typically written as follows: if v is a vector, and
R is a rotation matrix, then v’ is v rotated by R according to the formula:
v’=Rv
You can combine multiple rotation matrices reflecting multiple sequential rotations into
a single rotation matrix using usual matrix multiplication. If the rotation matrices are
expressed in terms of fixed, global coordinates, then they are combined as follows:
R c = Ri R2
Here R c is the combined rotation matrix reflecting a rotation first by R, and then by R 2 .
If the rotation matrices are expressed in terms of rotating, body-fixed coordinates, then
they are combined in the reverse order as follows:
R c = R2 Ri
We won’t go into the proof of this relation, but the reason it’s different depending on
how you’ve defined your rotation matrices is that rotation matrices defined in fixed
coordinates are unaffected by the rotation itself since the coordinate axes stay fixed. On
the other hand, if the rotation matrices are defined relative to a coordinate system that
is rotating due to sequential application of rotation matrices, then all rotation matrices
after the first will be affected since they were first defined relative to the original state
of the coordinate system—that is, before the first rotation matrix was applied. This
means that the subsequent rotation matrices must be corrected to reflect the new system
228 | Chapter 11: Rotation in 3D Rigid-Body Simulators
www.it-ebooks.info
as affected by the previous rotation before they can be correctly applied. In other words,
you have to rotate R, by R, to get a new R 2 before applying it. All this happens to work
out in such a way that you reverse the order of multiplication of rotation matrices when
they are defined in a rotating coordinate system.
Figure 11-1 shows a right-handed coordinate system that illustrates the directions of
positive rotation about each coordinate axis.
Let’s consider rotation around the z-axis where the point shown in Figure 11 -2 is rotated
through an angle 0.
The coordinates of the point before the rotation are (x,y,z) and after the rotation the
coordinates are (x r ,y r , z r ). The rotated coordinates are related to the original coordinates
and the rotation angle by the following:
x r = x cos 0 — y sin 0
y r = x sin 0 -)- y cos 0
z r = z
Rotation Matrices | 229
www.it-ebooks.info
Notice that since the point is rotating about the z-axis, its z coordinate remains un¬
changed. To write this in the vector-matrix notation, v’ = R v, let v = [x y z] and let R
be the matrix:
cos(0 z ) —sin(0 z ) 0
sin(0 z ) cos(# z ) 0
0 0 1
Here v’ will be the new, rotated vector, v’ = [x r y r zj.
Rotation about the x- and y-axes is similar to the z-axis; however, in those cases the x
andy coordinates remain constant during rotations about each axis, respectively. Look¬
ing at rotation about each axis separately will yield three rotation matrices similar to the
one we just showed you for rotation about the z-axis.
For rotation about the x-axis, the matrix is:
1 0 0
0 cos (# x ) —sin(0 x )
0 sin(d x ) cos (0 X )
And for rotation about the y-axis, the matrix is:
cos(d y ) 0 sin(0 y )
0 1 0
—sin(0y) 0 cos(0 y )
These are the rotation matrices you typically see in computer graphics texts in the con¬
text of matrix transforms, such as translation, scaling, and rotation. You can combine
all three of these matrices into a single rotation matrix to represent combinations of
rotations about each coordinate axis, using matrix multiplication as mentioned earlier.
In rigid-body simulations, you can use a rotation matrix to represent the orientation of
a rigid body. Another way to think of it is the rotation matrix, when applied to the
unrotated rigid body aligned with the fixed global coordinate system, will rotate the
rigid body’s coordinates so as to resemble the body’s current orientation at any given
time. This leads to another important consideration when using rotation matrices to
keep track of orientation in rigid-body simulations: the fact that the rotation matrix will
be a function of time.
Once you set up your initial rotation matrix for the rigid body, you’ll never directly
calculate it again from orientation angles; instead, the forces and moments applied to
the rigid body will change the body’s angular velocity, likewise causing small changes
230 | Chapter 11: Rotation in 3D Rigid-Body Simulators
www.it-ebooks.info
in orientation at each time step throughout the simulation. Thus, you can see that you
must have a means of relating the rotation matrix to angular velocity so that you can
update the orientation accordingly. The formula you need is as follows:
d R / dt = Q R
Here, fl is a skew symmetric matrix built from the angular velocity vector components
as follows:
fl =
0
<x > 7
—a> z 0)y
0 — (D x
w x 0
Notwithstanding a rigorous proof of this relation, it’s easy to see its beauty, which is that
you can differentiate the rotation matrix by simply matrix multiplying by the angular
velocity (in the form of fl). In a simulation you’ll know your initial rotation matrix, and
you’ll calculate the angular velocity at each time step; thus, you can easily progress, or
integrate, the rotation matrix.
You should be able to see here that since you’ll only explicitly calculate the rotation
matrix once and will update it with a matrix multiply, you won’t have to use computa¬
tionally expensive trigonometric functions during each time step. Further, you avoid
the singularity problem mentioned in the introduction to this chapter.
It should also be obvious that you gain these benefits at some price. First, you have to
deal with nine parameters in the rotation matrix (each element in the 3x3 rotation
matrix) to represent three angular degrees of freedom. Secondly, in order to do that,
you need to impose constraints on the rotation matrix; specifically, you need to enforce
the constraint that the matrix be orthogonal with a determinant of 1 so that it satisfies
the following (each column in the matrix represents a unit vector, and they are all at
right angles to each other): 1
r t r = i
Here R T is the transpose of R, and I is the identity matrix. Due to numerical errors such
as roundoff and truncation, you’ll have to enforce this constraint very often in your
simulation. Otherwise, your rotation matrix will do more than rotate your objects, it
may scale or translate them too.
Instead of dealing with nine parameters and trying to constrain six degrees of freedom
so that only the three you want can be represented, you could take an alternative ap-
1. Two vectors are orthogonal if their dot product is 0.
Rotation Matrices | 231
www.it-ebooks.info
proach that lets you keep the advantages rotation matrices have to offer, but at a cheaper
price. That alternative, quaternions, is the subject of the next section.
Quaternions
Quaternions are somewhat of a mathematical oddity. They were developed over 100
years ago by William Hamilton through his work in complex (imaginary) math but have
found very little practical use. A quaternion is a quantity, kind of like a vector, but made
up of four components. It is typically written in the form:
q = qo + q x * + q y j + q z k
A quaternion is really a four-dimensional quantity in complex space and, unfortunately,
does not lend itself to visualization. Don’t worry, though: our use of quaternions to
represent orientation in three dimensions does allow us to attach a physical meaning to
them, as you’ll see in a moment.
Of particular interest to us is what’s known as a unit quaternion that satisfies the fol¬
lowing:
qo 2 + q x 2 + q y 2 + q z 2 = i
This is analogous to a normalized, or unit, vector.
You can also write a quaternion in the form q = [q 0 , v], where v is the vector, q y i + q y )
+ q z k, and q„ is a scalar. In the context of rotation, v represents the direction in which
the axis of rotation points. For a given rotation, 0, about an arbitrary axis represented
by the unit vector u, the representative quaternion can be written as follows:
q = [cos(0/2), sin(0/2) u]
This is illustrated in Figure 11 -3 for an arbitrary rigid body rotating about an axis passing
through its center of gravity. The rigid body rotates through an angle 0 from the position
shown in light gray to the position shown in dark gray. Here, the unit vector u is the
vector v normalized to unit length.
232 | Chapter 11: Rotation in 3D Rigid-Body Simulators
www.it-ebooks.info
You can readily see that quaternions, when used to represent rotation or orientation,
require you to deal with only four parameters instead of nine, subject to the easily
satisfied constraint that the quaternion be a unit quaternion.
The use of quaternions to represent orientation is similar to how you would use rotation
matrices. First, you set up a quaternion that represents the initial orientation of the rigid
body at time 0 (this is the only time you’ll calculate the quaternion explicitly). Then you
update the orientation to reflect the new orientation at a given instant in time using the
angular velocities that are calculated for that instant. As you can see here, the differential
equation relating an orientation quaternion to angular velocity is very similar to that
for rotation matrices:
dq/dt = (1/2) © q
Here, the angular velocity is written in quaternion form as [0, to] and is expressed in
fixed, global coordinates, (to is still angular velocity, but you have to put it in quaternion
form instead of vector form when multiplying it by a quaternion q.) If to is expressed in
rotating, body-fixed coordinates, then you need to use this equation:
dq/dt = (1/2) q ©
As with rotation matrices, you can use quaternions to rotate points or vectors. If v is a
vector, then v’ is the rotated vector subject to the quaternion q:
v’= qvq*
Here q* is the conjugate of the quaternion q defined as:
q* = qo - q x i - q y j - q z k
You can also use the preceding formula to convert vectors from one coordinate system
to another, where one is rotated relative to the other. You have to do this, for example,
in your simulations where you are converting forces defined in fixed, global coordinates
Quaternions | 233
www.it-ebooks.info
to rotating, body-fixed coordinates so that you can apply the forces to the body; or you
might have to convert a body’s velocity defined in global coordinates to body coordinates
so that you can use the velocity in force calculations.
Quaternion Operations
As with vectors and matrices, quaternions have their own rules for the various opera¬
tions that you’ll need, such as multiplication, addition, subtraction, and so on. To make
it easy on you, we’ve included sample code in Appendix C that implements all of the
quaternion operations you’ll need; however, we want to highlight a few of the more
important ones here.
The Quaternion class is defined with a scalar component, n, and vector component, v,
where v is the vector, xi + yj + zk. The class has two constructors, one of which initializes
the quaternion to 0, and the other of which initializes the elements to those passed to
the constructor:
class Quaternion {
public:
float n; // number (scalar) part
Vector v; // vector part: v.x, v.y, v.z
Quaternion(void);
Quaternion(float e0, float el, float e2, float e3);
};
Magnitude
The Magnitude method returns the magnitude of the quaternion according to the fol¬
lowing formula:
This is similar to calculating the magnitude of a vector, except that for quaternions you
have to take the fourth term, the scalar n, into account.
Here’s the code that calculates the magnitude for our Quaternion class:
inline float Quaternion::Magnitude(void)
{
return (float) sqrt(n*n + v.x*v.x + v.y*v.y + v.z*v.z);
}
234 | Chapter 11: Rotation in 3D Rigid-Body Simulators
www.it-ebooks.info
Conjugate: The ~ operator
The conjugate of the product of quaternions is equal to the product of the quaternion
conjugates, but in reverse order:
~(qp) = (~p)(~q)
Here’s the code that computes the conjugate for our Quaternion class:
Quaternion operator-(void) const { return Quaternion( n.
-v.x,
-v.y.
-v.z);}
QVRotate
This function rotates the vector v by the unit quaternion q according to this formula:
p’ = (q)(v)(~q)
Here, ~q is the conjugate of the unit quaternion, q:
inline Vector QVRotate(Quaternion q, Vector v)
{
Quaternion t;
t = q*v*(~q);
return t.GetVectorQ;
}
This operator takes the conjugate of the quaternion, ~q, which is simply the negative of
the vector part. If q = [n, x i + y j + zk], then ~q = [n, (-x) i + (-y)) + (-z) k].
Quaternion multiplication: The * operator
This operator performs quaternion multiplication according to the following formula:
q p = n q n p - v q • v p + n q v p + n p v q + (v q X v p )
Here, n q n p - v q • v p is the scalar part of the result while n q v p + n p v q + (v q x v p ) is the
vector part. Also note that v q and v p are the vector parts of q and p, respectively, • is the
vector dot product operator, and x is the vector cross product operator.
Quaternion multiplication is associative but not commutative, thus:
q(ph) = (qp)h
qp^pq
Here’s the code that multiplies two Quaternions, ql and q2:
Quaternions | 235
www.it-ebooks.info
Inline Quaternion operator*(Quaternlon ql, Quaternion q2)
{
return Quaternlon(ql.n*q2.n - ql.v.x*q2.v.x
- ql.v.y*q2.v.y - ql.v.z*q2.v.z,
ql.n*q2.v.x + ql.v.x*q2.n
+ ql.v.y*q2.v.z - ql.v.z*q2.v.y,
ql.n*q2.v.y + ql.v.y*q2.n
+ ql.v.z*q2.v.x - ql.v.x*q2.v.z,
ql.n*q2.v.z + ql.v.z*q2.n
+ ql.v.x*q2.v.y - ql.v.y*q2.v.x);
}
Vector multiplication: The * operator
This operator multiplies the quaternion, q, by the vector v as though the vector v were
a quaternion with its scalar component equal to 0. There are two forms of this operator
depending on the order in which the quaternion and vector are encountered. Since v is
assumed to be a quaternion with its scalar part equal to 0, the rules of multiplication
follow those outlined earlier for quaternion multiplication:
Inline Quaternion operator*(Quaternlon q, Vector v)
{
return Quaternlon( -(q.v.x*v.x + q.v.y*v.y + q.v.z*v.z),
q.n*v.x + q.v.y*v.z - q.v.z*v.y,
q.n*v.y + q.v.z*v.x - q.v.x*v.z,
q.n*v.z + q.v.x*v.y - q.v.y*v.x);
}
Inline Quaternion operator*(Vector v, Quaternion q)
{
return Quaternlon( -(q.v.x*v.x + q.v.y*v.y + q.v.z*v.z),
q.n*v.x + q.v.z*v.y - q.v.y*v.z,
q.n*v.y + q.v.x*v.z - q.v.z*v.x,
q.n*v.z + q.v.y*v.x - q.v.x*v.y);
}
MakeQFromEulerAngles
This function constructs a quaternion from a set of Euler angles.
For a given set of Euler angles, yaw (t|/), pitch (x), and roll (<p), defining rotation about
the z-axis, then the y-axis, and then the x-axis, you can construct the representative
rotation quaternion. You do this by first constructing a quaternion for each Euler angle
and then multiplying the three quaternions following the rules of quaternion multipli¬
cation. Here are the three quaternions representing each Euler rotation angle:
qroii = [cos(cp/2), (sin(cp/2)) i + 0 j + 0 k]
qpitch = [cos(t 12), 0 i + (sin(x 12)) j + 0 k]
q yaw = [cos(\|/ 12), 0 i + 0 j + (sin(\|/ 12)) k]
236 | Chapter 11: Rotation in 3D Rigid-Body Simulators
www.it-ebooks.info
Each one of these quaternions is of unit length. 2
Now you can multiply these quaternions to obtain a single one that represents the ro¬
tation, or orientation, defined by the three Euler angles:
Q = by aw Q pi tell Qroll
Performing this multiplication yields:
q = [{cos(cp/2) cos(t 12) cos(q/ 12) + sin(cp/2) sin(x /2) sin(\p 12)},
{sin(cp/2) cos(t 12) cos(\|/ 12) - cos(cp/2) sin(x 12) sin(\|/ /2)} i +
{cos(cp/2) sin(x 12) cos(\|/ 12) + sin(cp/2) cos(x 12) sin(\|/ /2)} j +
{cos(cp/2) cos(x 12) sin(\|/ 12) - sin(cp/2) sin(x 12) cos(\|/ 12)} k]
Here’s the code that takes three Euler angles and returns a quaternion:
Inline Quaternion MakeQFromEulerAngles(float x, float y, float z)
{
Quaternion q;
double roll = DegreesToRadians(x);
double pitch = DegreesToRadians(y);
double yaw = DegreesToRadians(z);
double cyaw, cpitch, croll, syaw, spitch, sroll;
double cyawcpitch, syawspitch, cyawspitch, syawcpitch;
cyaw = cos(0.5f * yaw);
cpitch = cos(0.5f * pitch);
croll = cos(0.5f * roll);
syaw = sin(0.5f * yaw);
spitch = sin(0.5f * pitch);
sroll = sin(0.5f * roll);
cyawcpitch = cyaw*cpitch;
syawspitch = syaw*spitch;
cyawspitch = cyaw*spitch;
syawcpitch = syaw*cpitch;
q.n = (float) (cyawcpitch * croll + syawspitch * sroll);
q.v.x = (float) (cyawcpitch * sroll - syawspitch * croll);
q.v.y = (float) (cyawspitch * croll + syawcpitch * sroll);
q.v.z = (float) (syawcpitch * croll - cyawspitch * sroll);
return q;
}
2. You can verify this by recalling the trigonometric relation cos 2 0 + sin 2 0=1.
Quaternions | 237
www.it-ebooks.info
MakeEulerAnglesFromQ
This function extracts the three Euler angles from a given quaternion.
You can extract the three Euler angles from a quaternion by first converting the qua¬
ternion to a rotation matrix and then extracting the Euler angles from the rotation
matrix. Let R be a nine-element rotation matrix:
r n
r 12
t*21
?22
r 23
Til
1*32
r 33
and let q be a quaternion:
q = [n, x i + y j + z k]
Then each element in R is calculated from q as follows:
9999
r u = n- + x z - y z - z z
i 2 i = 2xy+2zn
r 3 i = 2zx — 2yn
r \2 = 2xy — 2zn
r 22 = n 2 - x 2 + y 2 - z 2
r 3 2 = 2zy + 2xn
rj 3 = 2xz + 2yn
r 23 = 2yz — 2xn
r 33 = n- - x- - y z + z z
To extract the Euler angles, yaw (\|/), pitch (t), and roll (cp), from R, you can use these
relations:
tan i|/ = r 2 i / ru
sin t = -r 3 i
tan cp = r 32 / r 33
Here’s the code that extracts the three Euler angles, returned in the form of a Vector,
from a given quaternion:
inline Vector MakeEulerAnglesFromQ(Quaternion q)
{
double rll, r21, r31, r32, r33, rl2, rl3;
double q00, qll, q22, q33;
double tnp;
Vector u;
238 | Chapter 11: Rotation in 3D Rigid-Body Simulators
www.it-ebooks.info
q00 = q.n * q.n;
qll = q.v.x * q.v.x;
q22 = q.v.y * q.v.y;
q33 = q.v.z * q.v.z;
rll = q00 + qll - q22 - q33;
r21 = 2 * (q.v.x*q.v.y + q.n*q.v.z);
r31 = 2 * (q.v.x*q.v.z - q.n*q.v.y);
r32 = 2 * (q.v.y*q.v.z + q.n*q.v.x);
r33 = q00 - qll - q22 + q33;
tnp = fabs(r31);
if (trip > 0.999999)
{
rl2 = 2 * (q.v.x*q.v.y - q.n*q.v.z);
rl3 = 2 * (q.v.x*q.v.z + q.n*q.v.y);
u.x = RadlansToDegrees(0.0f); //roll
u.y = RadiansToDegrees((float) (-(pl/2) * r31/tnp)); // pitch
u.z = RadiansToDegrees((float) atan2(-rl2, -r31*rl3)); // yaw
return u;
}
u.x = RadiansToDegrees((float) atan2(r32, r33)); // roll
u.y = RadiansToDegrees((float) asin(-r31)); // pitch
u.z = RadiansToDegrees((float) atan2(r21, rll)); // yaw
return u;
}
Quaternions in 3D Simulators
The quaternion operations just presented are required when you are using quaternions
to represent orientation in 3D simulations. All the 3D simulations discussed in this book
use these quaternion operations, and in this section we’ll highlight where they are used
in the context of the airplane example presented in Chapter 15.
When initializing the orientation of the airplane, you have to set its orientation qua¬
ternion to something corresponding to the Euler angles you desire. You do so as follows:
Airplane.qOrientatlon = MakeQFromEulerAngles(iRoll, IPitch, iYaw);
In this code sample, Airplane is a rigid-body class with the property qOrientation,
which represents the orientation quaternion, which is a Quaternion class. iRoll,
iPitch, and iYaw are the three Euler angles describing the orientation of the airplane.
If at any time you want to report the Euler angles—for example, in a heads-up display¬
like interface for the game player—you can use MakeEulerAnglesFrornQ, as follows:
Quaternions in 3D Simulators | 239
www.it-ebooks.info
// get the Euler angles for our information
Vector u;
u = MakeEulerAnglesFromQ(Airplane.qOrientation);
Airplane.vEulerAngles.x = u.x; // roll
Airplane.vEulerAngles.y = u.y; // pitch
Airplane.vEulerAngles.z = u.z; // yaw
Very often, it’s more convenient to calculate loads on an object like the airplane using
body-fixed coordinates. For example, when computing aerodynamic drag on the air¬
plane, you’ll want to know the relative air velocity over the aircraft in body-fixed coor¬
dinates. The resulting drag force will also be in body-fixed coordinates. However, when
resolving all the loads on the aircraft to determine its motion in earth-fixed coordinates,
you’ll want to convert those forces from body-fixed coordinates to earth-fixed coordi¬
nates. You can use QVRotate to rotate any vector, such as a force vector, from one co¬
ordinate system to another. The following code sample shows how QVRotate is used to
convert a force vector in body-fixed coordinates to the equivalent force in earth-fixed
coordinates.
void CalcAirplaneLoads(vold)
{
// Convert forces from model space to earth space
Airplane.vForces = QVRotate(Airplane.qOrlentation, Fb);
}
Throughout the simulation, you’ll have to update the airplane’s orientation by integrat¬
ing the angular equations of motion. The first step in handling angular motion is to
calculate the new angular velocity at a given time step based on the previously calculated
moments acting on the airplane and its mass properties. We do this in body coordinates
using the angular equation of motion:
£ M cg = dH cg /dt = I (dco/dt) + (to X (I to))
The next step is to integrate again to update the airplane’s orientation, which is expressed
as a quaternion. Here, you need to use the differential equation relating an orientation
quaternion to angular velocity that we discussed earlier:
dq/dt = (1/2) to q
240 | Chapter 11: Rotation in 3D Rigid-Body Simulators
www.it-ebooks.info
Next, to enforce the constraint that this orientation quaternion be a unit quaternion,
you must normalize the orientation quaternion. The following code sample illustrates
these steps:
// calculate the angular velocity of the airplane in body space:
Airplane.vAngularVelocity += Airplane.nlnertialnverse *
(Airplane.vMoments -
(Airplane. vAngularVelocity''
(Airplane.mlnertia *
Airplane.vAngularVelocity)))
* dt;
// calculate the new rotation quaternion:
Airplane.qOrientation += (Airplane.qOrientation *
Airplane.vAngularVelocity) *
(0.5f * dt);
// now normalize the orientation quaternion:
mag = Airplane. qOrientation. MagnitudeQ;
if (mag != 0)
Airplane.qOrientation /= mag;
// calculate the velocity in body space:
// (we'll need this to calculate lift and drag forces)
Airplane.vVelocityBody = QVRotate(~Airplane.qOrientation,
Airplane.vVelocity);
Notice the last line of code in the preceding sample. That line converts the airplanes
velocity vector from earth-fixed coordinates to body-fixed coordinates using QVRo
tate. Recall that it’s more convenient to compute body forces in body-fixed coordinates.
QVRot a t e allows you to work with vectors back and forth from body-fixed to earth-fixed
coordinates.
Quaternions in 3D Simulators | 241
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 12
3D Rigid-Body Simulator
In this chapter we’ll show you how to make the leap from 2D to 3D by implementing a
rigid-body simulation of an airplane. Specifically, this is a simulation of the hypothetical
airplane model that we’ll discuss extensively in Chapter 15. This airplane is of typical
configuration with its large wings forward, its elevators aft, a single vertical tail, and
plain flaps fitted on the wings.
As with the 2D simulator in previous chapters, we’ll concentrate on the code that im¬
plements the physics part of the simulator and not the platform-specific GUI aspects of
the simulations.
As in 2D, there are four main elements to this 3D simulation—the model, integrator,
user input, and rendering. Remember, the model refers to your idealization of the thing
—an airplane, in this case—that you are trying to simulate, while the integrator refers
to the method by which you integrate the differential equations of motion. These two
elements take care of most of the physics of the simulation. The user input and rendering
elements refer to how you’ll allow the user to interact with and view your simulation.
In this simulation, the world coordinate system has its positive x-axis pointing into the
screen, its positive y-axis pointing to the left of your screen, and the positive z-axis
pointing up. Also, the local, or body-fixed, coordinate system has its positive x-axis
pointing toward the front of the airplane, its positive y-axis pointing to the port side
(left side), and its positive z-axis pointing up. Since this is a 3D simulation of an airplane,
once you get it running, you’ll be able to fly in any direction, looping, banking, diving,
and climbing, or performing any other aerobatic maneuver you desire.
Model
One of the most important aspects of this simulation is the flight model. We’ll spend all
of Chapter 15 discussing the physics behind this flight model, so we won’t include that
discussion here except to introduce a few key bits of code.
243
www.it-ebooks.info
To implement the flight model, you first need to prepare a rigid-body structure to en¬
capsulate all of the data required to completely define the state of the rigid body at any
instant during the simulation. We’ve defined a structure called RigidBody for this pur¬
pose:
typedef struct _RtgidBody {
float
fMass;
//
total mass
Matrix3x3
mlnertia;
//
mass moment of inertia
//
in body coordinates
Matrix3x3
mlnertialnverse;
//
inverse of mass moment of inertia
Vector
vPosition;
//
position in earth coordinates
Vector
vVelocity;
//
velocity in earth coordinates
Vector
vVelocityBody;
//
velocity in body coordinates
Vector
vAngularVelocity;//
angular velocity in body coordinates
Vector
vEulerAngles;
//
Euler angles in body coordinates
float
fSpeed;
//
speed (magnitude of the velocity)
Quaternion
qOrientation;
//
orientation in earth coordinates
Vector
vForces;
//
total force on body
Vector
vMoments;
//
total moment (torque) on body
} RigidBody, *pRigidBody;
You’ll notice that it is very similar to the RigidBody2D structure that we used in the 2D
hovercraft simulation. One significant difference, however, is that in the 2D case, ori¬
entation was a single float value, and now in 3D it’s a quaternion of type Quaternion.
We discussed the use of quaternions for tracking rigid-body orientation in the previous
chapter, and Appendix C contains a complete definition of the Quaternion class.
The next step in defining the flight model is to prepare an initialization function to
initialize the airplane at the start of the simulation. For this purpose, we’ve prepared a
function called InltlallzeAlrplane:
RigidBody Airplane; // global variable representing the airplane
void InitializeAirplane(void)
{
float iRoll, iPitch, iYaw;
// Set initial position
Airplane.vPosition.x = -5000.0f;
Airplane.vPosition.y = 0.0f;
Airplane.vPosition.z = 2000.0f;
// Set initial velocity
Airplane.vVelocity.x = 60.0f;
Airplane.vVelocity.y = 0.0f;
Airplane.vVelocity.z = 0.0f;
244 | Chapter 12:3D Rigid-Body Simulator
www.it-ebooks.info
Airplane.fSpeed = 60.0f;
// Set initial angular velocity
Airplane.vAngularVelocity.x = 0.0f;
Airplane.vAngularVelocity.y = 0.0f;
Airplane.vAngularVelocity.z = 0.0f;
// Set the initial thrust, forces, and moments
Airplane.vForces.x = 500.0f;
Airplane.vForces.y = 0.0f;
Airplane.vForces.z = 0.0f;
ThrustForce = 500.0;
Airplane.vMoments.x = 0.0f;
Airplane.vMoments.y = 0.0f;
Airplane.vMoments.z = 0.0f;
// Zero the velocity in body space coordinates
Airplane.vVelocityBody.x = 0.0f;
Airplane.vVelocityBody.y = 0.0f;
Airplane.vVelocityBody.z = 0.0f;
// Set these to false at first,
// you can control later using the keyboard
Stalling = false;
Flaps = false;
// Set the initial orientation
iRoll = 0.0f;
iPitch = 0.0f;
iYaw = 0.0f;
Airplane.qOrientation = MakeQFromEulerAngles(iRoll, iPitch, iYaw);
// Now go ahead and calculate the plane's mass properties
CalcAirplaneMassProperties();
}
This function sets the initial location, speed, attitude, and thrust for the airplane and
goes on to calculate its mass properties by making a call to CalcAirplaneMassProper
ties. You’ll see much more of this function in Chapter 15, so we won’t show the whole
thing here. We do want to point out a portion of the code that is distinctly different from
what you do in a 2D simulation, and that’s the calculation of the moment of inertia
tensor:
void CalcAirplaneMassProperties(void)
{
// Now calculate the moments and products of inertia for the
// combined elements.
Model | 245
www.it-ebooks.info
// (This
Ixx = 0;
Ixy = 0;
for (i =
{
Ixx
Iyy
Izz
Ixy
Ixz
lyz
}
inertia matrix (tensor) is in body coordinates)
Iyy = 0; Izz = 0;
Ixz = 0; lyz = 0;
0; i< 8; i++)
+= Element[i].vLocallnertia.x + Elementfi].fMass *
(Element[i].vCGCoords.y*Element[i].vCGCoords.y +
Element[i].vCGCoords.z*Element[i].vCGCoords.z);
+= Element[i].vLocallnertia.y + Element[i].fMass *
(Element[i].vCGCoords.z*Element[l].vCGCoords.z +
Element[i].vCGCoords.x*Element[i].vCGCoords.x);
+= Element[i].vLocallnertia.z + Elementfi].fMass *
(Element[i].vCGCoords.x*Element[i].vCGCoords.x +
Element[i].vCGCoords.y*Element[i].vCGCoords.y);
+= Element[i].fMass * (Elementfi].vCGCoords.x *
Element[i].vCGCoords.y);
+= Element[i].fMass * (Element[i].vCGCoords.x *
Element[i].vCGCoords.z);
+= Element[i].fMass * (Element[i].vCGCoords.y *
Element[i].vCGCoords.z);
// Finally set up the airplane's mass
// inverse of the inertia matrix
Airplane.fMass = mass;
Airplane.mlnertia.ell = Ixx;
Airplane.mlnertia.el2 = -Ixy;
Airplane.mlnertia.el3 = -Ixz;
Airplane.mlnertia.e21 = -Ixy;
Airplane.mlnertia.e22 = Iyy;
Airplane.mlnertia.e23 = -lyz;
Airplane.mlnertia.e31 = -Ixz;
Airplane.mlnertia.e32 = -lyz;
Airplane.mlnertia.e33 = Izz;
and its inertia matrix and take the
Airplane.mlnertialnverse = Airplane.mlnertia.Inverse();
}
The airplane is modeled by a number of elements, each representing a different part of
the airplane’s structure—for example, the tail rudder, elevators, wings, and fuselage (see
Chapter 15 for more details). The code specified here takes the mass properties of each
element and combines them, using the techniques discussed in Chapter 7 to come up
with the combined inertia tensor for the entire aircraft. The important distinction be¬
tween these calculations in a 3D simulation and the 2D simulation is that here the inertia
is a tensor and in 2D it is a single scalar.
InitializeAirplane is called at the very start of the program. We found it convenient
to make the call right after the applications main window is created.
The final part of the flight model has to do with calculating the forces and moments
that act on the airplane at any given instant in time during the simulation. As in the 2D
246 | Chapter 12:3D Rigid-Body Simulator
www.it-ebooks.info
hovercraft simulation, without this sort of function, the airplane will do nothing. For
this purpose we’ve defined a function called CalcAirpla neLoad s, which is called at every
step through the simulation. This function relies on a couple of other functions—
namely, LiftCoefficient, DragCoefficient, RudderLiftCoefficient, and Rudder
DragCoef f icient. All of these functions are shown and discussed in detail in the section
“Modeling” on page 305 in Chapter 15.
For the most part, the code contained in CalcAirplaneLoads is similar to the code you’ve
seen in the CalcLoads function of the hovercraft simulation. CalcAirplanLoads is a
little more involved since the airplane is modeled by a number of elements that con¬
tribute to the total lift and drag on the airplane. There’s also another difference that
we’ve noted here:
void CalcAtrptaneLoads(vold)
{
// Convert forces from node! space to earth space
Airplane.vForces = QVRotate(Airplane.qOrientation, Fb);
// Apply gravity (g is defined as -32.174 ft/s A 2)
Airplane.vForces.z += g * Airplane.fMass;
}
Just about all of the forces acting on the airplane are first calculated in body-fixed co¬
ordinates and then converted to earth-fixed coordinates before the gravity force is ap¬
plied. The coordinate conversion is effected through the use of the function QVRotate,
which rotates the force vector based on the airplane’s current orientation, represented
by a quaternion. 1
Integration
Now that the code to define, initialize, and calculate loads on the airplane is complete,
you need to develop the code to actually integrate the equations of motion so that the
simulation can progress through time. The first thing you need to do is decide on the
integration scheme that you want to use. In this example, we decided to go with the
basic Euler’s method. We’ve already discussed some better methods in Chapter 7. We’re
going with Euler’s method here because it’s simple and we didn’t want to make the code
1. QVRotate is defined in Appendix C.
Integration | 247
www.it-ebooks.info
here overly complex, burying some key code that we need to point out to you. In practice,
you’re better off using one of the other methods we discuss in Chapter 7 instead of Euler’s
method. With that said, we’ve prepared a function called StepSimulation that handles
all of the integration necessary to actually propagate the simulation:
void StepSimulation(float dt)
{
// Take care of translation first:
// (If this body were a particle, this is all you would need to do.)
Vector Ae;
// calculate all of the forces and moments on the airplane:
CalcAirplaneLoads();
// calculate the acceleration of the airplane in earth space:
Ae = Airplane.vForces / Airplane.fMass;
// calculate the velocity of the airplane in earth space:
Airplane.vVelocity += Ae * dt;
// calculate the position of the airplane in earth space:
Airplane.vPosition += Airplane.vVelocity * dt;
// Now handle the rotations:
float mag;
// calculate the angular velocity of the airplane in body space:
Airplane.vAngularVelocity += Airplane.mlnertialnverse *
(Airplane.vMoments -
(Airplane.vAngularVelocity A
(Airplane.mlnertia *
Airplane.vAngularVelocity)))
* dt;
// calculate the new rotation quaternion:
Airplane.qOrientation += (Airplane.qOrientation *
Airplane.vAngularVelocity) *
(0.5f * dt);
// now normalize the orientation quaternion:
mag = Airplane. qOrientation. MagnitudeQ;
if (mag != 0)
Airplane.qOrientation /= mag;
// calculate the velocity in body space:
// (we'll need this to calculate lift and drag forces)
Airplane.vVelocityBody = QVRotate(~Airplane.qOrientation,
Airplane.vVelocity);
// calculate the air speed:
248 | Chapter 12:3D Rigid-Body Simulator
www.it-ebooks.info
Airplane.fSpeed = Airplane.vVelocity.Magnitude();
// get the Euler angles for our information
Vector u;
u = MakeEulerAnglesFromQ(Airplane.qOrientation);
Airplane.vEulerAngles.x = u.x; // roll
Airplane.vEulerAngles.y = u.y; // pitch
Airplane.vEulerAngles.z = u.z; // yaw
}
The very first thing that StepSimulation does is call CalcAirplaneLoads to calculate
the loads acting on the airplane at the current instant in time. StepSimulation then
goes on to calculate the linear acceleration of the airplane based on current loads. Next,
the function goes on to integrate, using Eulers method, once to calculate the airplane’s
linear velocity and then a second time to calculate the airplanes position. As we’ve
commented in the code, if you were simulating a particle this is all you would have to
do; however, since this is not a particle, you need to handle angular motion.
The first step in handling angular motion is to calculate the new angular velocity at this
time step, using Euler integration, based on the previously calculated moments acting
on the airplane and its mass properties. We do this in body coordinates using the fol¬
lowing equation of angular motion but rewritten to solve for do:
£ M cg = dH cg /dt = I (dw/dt) + (to x (I to))
The next step is to integrate again to update the airplane’s orientation, which is expressed
as a quaternion. Here, you need to use the differential equation relating an orientation
quaternion to angular velocity that we showed you in Chapter 11:
dq/dt = (1/2) to q
Next, to enforce the constraint that this orientation quaternion be a unit quaternion,
the function goes ahead and normalizes the orientation quaternion.
Since the linear velocity was previously calculated in global coordinates (the fixed co¬
ordinate system), and since CalcAirplaneLoads needs the velocity in the body-fixed
(rotating) coordinates system, the function goes ahead and rotates the velocity vector,
storing the body-fixed vector in the vVelocityBody member of the RigidBody structure.
This is done here as a matter of convenience and uses the quaternion rotation function
QVRotate to rotate the vector based on the airplane’s current orientation. Notice here
that we use the conjugate of the orientation quaternion since we’re now rotating from
global coordinates to body coordinates.
Integration | 249
www.it-ebooks.info
As another convenience, we calculate the air speed, which is simply the magnitude of
the linear velocity vector. This is used to report the air speed in the main window title
bar.
Lastly, the three Euler angles—roll, pitch, and yaw—are extracted from the orientation
quaternion so that they can also be reported in the main window title bar. The function
to use here is MakeEulerAnglesFropiQ, which is defined in Appendix C.
Don’t forget, StepSimulation must be called once per simulation cycle.
Flight Controls
At this point, the simulation still won’t work very well because you have not implemented
the flight controls. The flight controls allow you to interact with the airplane’s various
controls surfaces in order to actually fly the plane. We’ll use the keyboard as the main
input device for the flight controls. Remember, in a physics-based simulation such as
this one, you don’t directly control the motion of the airplane; you control only how
various forces are applied to the airplane, which then, by integration over time, affect
the airplane’s motion.
For this simulation, the flight stick is simulated by the arrow keys. The down arrow pulls
back on the stick, raising the nose; the up arrow pushes the stick forward, causing the
nose to dive; the left arrow rolls the plane to the left (port side); and the right arrow
rolls the plane to the right (starboard side). The X key applies left rudder action to cause
the nose of the plane to yaw toward the left, while the C key applies right rudder action
to cause the nose to yaw toward the right. Thrust is controlled by the A and Z keys. The
A key increments the propeller thrust by 100 pounds, and the Z key decrements the
thrust by 100 pounds. The minimum thrust is 0, while the maximum available thrust is
3,000 pounds. The F key activates the landing flaps to increase lift at low speed, while
the D key deactivates the landing flaps.
We control pitch by deflecting the flaps on the aft elevators; for example, to pitch the
nose up, we deflect the aft elevator flaps upward (that is, the trailing edge of the elevator
is raised with respect to the leading edge). We control roll in this simulation by applying
the flaps differentially; for example, to roll right, we deflect the right flap upward and
the left flap downward. Finally, we control yaw by deflecting the vertical tail rudder; for
example, to yaw left, we deflect the trailing edge of the tail rudder toward the left.
We’ve prepared several functions to handle the flight controls that should be called
whenever the user is pressing one of the flight control keys. There are two functions for
the propeller thrust:
void IncThrust(vold)
{
ThrustForce += _DTHRUST;
if(ThrustForce > _MAXTHRUST)
250 | Chapter 12:3D Rigid-Body Simulator
www.it-ebooks.info
ThrustForce = _MAXTHRUST;
}
void DecThrust(void)
{
ThrustForce -= _DTHRUST;
if(ThrustForce < 0)
ThrustForce = 0;
}
IncThrust simply increases the thrust by _DTHRUST checking to make sure it does not
exceed _MAXTHRUST. We’ve defined _DTHRUST and _MAXTHRUST as follows:
#define _DTHRUST 100.0f
#define _MAXTHRUST 3000.0f
DecTh rust, on the other hand, decreases the thrust by _DTHRUST checking to make sure
it does not fall below 0.
To control yaw, we’ve prepared three functions that manipulate the rudder:
void LeftRudder(void)
{
Element[6].flncidence = 16;
}
void RightRudder(void)
{
Element[6].flncidence = -16;
}
void ZeroRudder(void)
{
Element[6].flncidence = 0;
}
LeftRudder changes the incidence angle of Element[6], the vertical tail rudder, to 16
degrees, while RightRudder changes the incidence angle to -16 degrees. ZeroRudder
centers the rudder at 0 degrees.
The ailerons, or flaps, are manipulated by these functions to control roll:
void RollLeft(void)
{
Element[0].iFlap = 1;
Element[3].iFlap = -1;
}
void RollRight(void)
{
Element[0].iFlap = -1;
Element[3].iFlap = 1;
}
Flight Controls | 251
www.it-ebooks.info
void ZeroAilerons(void)
{
Etement[0].iFtap = 0;
Element[3].iFlap = 0;
}
RollLeft deflects the port aileron, located on the port wing section (Element[0]), up¬
ward, and the starboard aileron, located on the starboard wing section (Element[3]),
downward. RollRight does just the opposite, and ZeroAilerons resets the flaps back
to their undeflected positions.
We’ve defined yet another set of functions to control the aft elevators so as to control
pitch:
void PitchUp(void)
{
Element[4].iFtap = 1;
Etement[5].iFtap = 1;
}
void PitchDown(void)
{
Element[4].iFtap = -1;
ElementfS].iFtap = -1;
}
void ZeroEtevators(void)
{
Etement[4].iFtap = 0;
Etement[5].iFtap = 0;
1
Element[4] and Element[5] are the elevators. Pitchllp deflects their flaps upward, and
PitchDown deflects their flaps downward. ZeroElevators resets their flaps back to their
undeflected positions.
Finally, there are two more functions to control the landing flaps:
void FtapsDown(void)
f
Etementfl].iFtap = - 1 ;
Etement[2].iFtap = -1;
Ftaps = true;
1
void ZeroFtaps(void)
{
Etementfl].iFtap = 0;
Etement[2].iFtap = 0;
252 | Chapter 12:3D Rigid-Body Simulator
www.it-ebooks.info
Flaps = false;
}
The landing flaps are fitted on the inboard wings sections, port and starboard, which
are Element[l] and Element[2]. FlapsDown deflects the flaps downward, while Zero
Flaps resets them back to their undeflected position.
As we said, these functions should be called when the user is pressing the flight control
keys. Further, they need to be called before StepSimulation is called so that they can
be included in the current time step’s force and moment calculations. The sequence of
calls should look something like this:
ZeroRudderQ;
ZeroAilerons();
ZeroElevators();
// pitch down
if (IsKeyDown(VKJJP))
PitchDown();
// pitch up
if (IsKeyDown(VK_DOWN))
PitchLlp();
// roll left
if (IsKeyDown(VK_LEFT))
RollLeft();
// roll right
if (IsKeyDown(VK_RIGHT))
RollRight();
// Increase thrust
if (IsKeyDown(0x41)) // A
IncThrust();
// Decrease thrust
if (IsKeyDown(0x5A)) // Z
DecThrust();
// yaw left
if (IsKeyDown(0x58)) // x
LeftRudder();
// yaw right
if (IsKeyDown(0x43)) // c
RightRudder();
Flight Controls | 253
www.it-ebooks.info
// landing flaps down
if (IsKeyDown(0x46)) //f
FlapsDownQ;
// landing flaps up
if (IsKeyDown(0x44)) // d
ZeroFlapsQ;
StepSinulation(dt);
Before StepSimulation is called, we check each of the flight control keys to see if it is
being pressed. If so, then the appropriate function is called.
The function IsKeyDown, which checks whether a certain key is pressed, looks like this
in a Windows implementation:
BOOL IsKeyDown(short KeyCode)
{
SHORT retval;
retval = GetAsyncKeyState(KeyCode);
if (HIBYTE(retval))
return TRUE;
return FALSE;
}
The important thing to note here is that the keys are being checked asynchronously
because it is possible that more than one key will be pressed at any given time, and they
must be handled simultaneously instead of one at a time (as would be the case in the
standard Windows message processing function).
The addition of flight control code pretty much completes the physics part of the sim¬
ulation. So far, you have the model, the integrator, and the user input or flight control
elements completed. All that remains is setting up the application’s main window and
actually drawing something that represents what you’re simulating. We’ll leave that part
up to you, or you can look at the example we’ve included on the book’s website to see
what we did on a Windows machine.
254 | Chapter 12:3D Rigid-Body Simulator
www.it-ebooks.info
CHAPTER 13
Connecting Objects
Simulating particles and rigid bodies is great fun, and with these simple entities you can
achieve a wide variety of effects or simulate a wide variety of objects. In this chapter
we’ll take things a step further, showing you how to simulate connected particles and
rigid bodies. Doing so opens a whole new realm of possibilities. In this book’s first
edition, David showed you how to use springs and particles to simulate cloth. Chap¬
ter 17 in the first edition covers that, and the corresponding “Cloth Simulation” example
on the book’s website implements the model. As shown in Figure 13-1, the flag model
is simply a collection of particles initially laid out in a grid pattern connected by linear
springs that are then rendered to look like cloth. The springs give structure to the par¬
ticles, keeping them organized in a mesh that can be rendered while allowing them to
move, emulating the movement of a flowing fabric.
255
www.it-ebooks.info
Figure 13-1. Network of particles and springs
Each line in the wireframe flag shown in Figure 13-1 represents a spring-damper ele¬
ment, while the nodes where these springs intersect represent the particles. We modeled
the springs using the spring-damper formulas that we showed you back in Chapter 3.
The (initially) horizontal and vertical springs provide the basic structure for the flag,
while the diagonal springs are there to resist shear forces and lend further strength to
the cloth. Without these shear springs, the cloth would be quite stretchy. Note that there
are no particles located at the intersection of the diagonal springs.
In this chapter, we’ll show you how to use those same techniques to simulate something
like a hanging rope or vine. You can use these techniques to simulate all sorts of things
besides cloth and rope or vines. For example, you can model the swing of a golf club if
you can imagine one rigid body representing the arm and another representing the golf
club. We’ll get to that example in Chapter 19, but for now let’s see how to model a hanging
rope or vine and some other springy objects.
256 | Chapter 13: Connecting Objects
www.it-ebooks.info
Application of linear springs is not the only method available to connect objects, but it
has the advantages of being conceptually simple, easy to implement, and effective. One
of the potential disadvantages is that you can run into numerical stability problems if
the springs are too stiff. We’ll talkmore about these issues throughout this chapter. Also,
the examples we’ll cover are in 2D for simplicity, but the techniques apply in 3D, too.
Springs and Dampers
You learned in Chapter 3 that springs are structural elements that, when connected
between two objects, apply equal and opposite forces to each object. This spring force
follows Hooke’s law and is a function of the stretched or compressed length of the spring
relative to the rest length of the spring and the spring constant. The spring constant is
a quantity that relates the force exerted by the spring to its deflection:
F s = -k s (L-r)
Here, F s is the spring force, k s is the spring constant, L is the stretched or compressed
length of the spring, and r is the rest length of the spring. The negative sign in the
preceding equation just means that the force is in the opposite direction of the dis¬
placement.
Dampers are usually used in conjunction with springs in numerical simulations. They
act like viscous drag in that dampers act against velocity. The force developed by a
damper is proportional to the relative velocity of the connected objects and a damping
constant, k d , that relates relative velocity to damping force.
Fd = -kd (vi - V 2 )
This equation shows the damping force, F d , as a function of the damping constant and
the relative velocity of the connected points on the two connected bodies.
Typically, springs and dampers are combined into a single spring-damper element
where a single formula is used to represent the combined force. In vector notation, the
formula for a spring-damper element connecting two bodies is:
F] = -{k s (L - r) + k d (( Vl - v 2 ) • L)/L} L/L
Here, is the force exerted on body 1, while the force, F 2 , exerted on body 2 is:
F 2 = -Ft
L is the length of the spring-damper (L, not in bold print, is the magnitude of the vector
L), which is equal to the vector difference in position between the connected points on
bodies 1 and 2. If the connected objects are particles, then L is equal to the position of
Springs and Dampers | 257
www.it-ebooks.info
body 1 minus the position of body 2. Similarly, and v 2 are the velocities of the con¬
nected points on bodies 1 and 2. The quantity (v t - v 2 ) represents the relative velocity
between the connected bodies.
It’s fairly straightforward to connect particles with springs (and dampers); you need
only specify the particles to which the spring is connected and compute the stretched
or compressed length of the spring as the particles move relative to each other. The force
generated by the spring is then applied equally (but in opposite directions) to the con¬
nected particles. This is a linear force.
For rigid bodies, things are a bit more complicated. First, not only do you have to specify
to which body the spring is attached, but you must also specify the precise points on
each object where the spring attaches. Then, in addition to the linear force applied by
the spring to each body, you must also compute the resulting moment on each body
causing each to rotate.
Connecting Particles
From swinging vines in Activision’s Pitfall to barnacle tongues in Valve Corporation’s
Half-Life, dangling rope-like objects have appeared in video games in various incarna¬
tions since the very early days of video gaming. Some implementations, such as those
in the 1982 versions of Pitfall, are implemented rather simply and unrealistically, while
others, such as barnacle tongues, are implemented more realistically in how they dangle
and swing. Whether it’s a vine, rope, chain, or tongue, you can use particles and springs
to simulate realistic rope-like behavior. We’ll show you how in the following simple
example.
Rope
You know from your real-life experience that ropes are flexible, although some are more
flexible than others. Ropes are elastic and stretch to varying extents. They drape when
suspended by their two ends. They bend when swinging or when collapsing on the
ground. We can capture all these behaviors using simple particles connected with
springs. Figure 13-2 illustrates the rope example we’ll cover here.
258 | Chapter 13: Connecting Objects
www.it-ebooks.info
Figure 13-2. Swinging rope
The example consists of a rope comprising 10 particles and 9 springs. At the start of the
simulation, the rope, originally extended straight out to the right, falls under the influ¬
ence of gravity, swinging left and right until it comes to rest (hanging straight down).
The dots represent particles and the lines represent springs. The topmost particle is
fixed, and the illustration on the left in Figure 13-2 shows the rope swinging down from
right to left while the illustration on the right shows the rope swinging back from left
to right.
This example uses all the same code and techniques presented in Chapter 7 through
Chapter 9 for simulating particles and rigid bodies. Really, the only difference is that we
have to compute a new force—the spring force on each object. But before we do that,
we have to define and initialize the springs.
Spring structure and variables
The following code sample shows the spring data structure we set up to store each
spring’s information:
typedef struct _Sprtng {
tnt Endl;
tnt End2;
float k;
float d;
Connecting Particles | 259
www.it-ebooks.info
float InitialLength;
} Spring, *pSpring;
Specifically, this information includes:
Endl
A reference to the first particle to which the spring is connected
End2
A reference to the second particle to which the spring is connected
k
The spring constant
d
The damping constant
InitialLength
The unstretched length of the spring
This structure is appropriate for connecting particles. We’ll make a slight modification
to this structure later, when we get to the example where we’re connecting rigid bodies.
There are defines and variables unique to this example that must be set up as follows:
#define
_NUM_OBJECTS
10
#define
_NUM_SPRINGS
9
#define
_SPRING_K
1000
#define
_SPRING_D
100
Particle
Objects[_NUM_0B3ECTS];
Spring
Springs[_NUM_SPRINGS];
As stated earlier, there are 10 particles (objects) and 9 springs in this simulation. The
arrays Objects and Springs are used to keep track of them. We also set up a few
defines representing the spring and damping constants. The values shown here are
arbitrary, and you can change them to suit whatever behavior you desire. The higher
the spring constant, the stiffer the springs; whereas the lower the spring constant, the
stretchier the springs. Stretchy springs make your rope more elastic. Keep in mind while
tuning these values that if you make the spring constant too high, you’ll probably have
to make the simulation time step smaller and/or use a robust integration scheme to
avoid numerical instabilities.
The damping constant controls how quickly the springiness of the springs dampens
out. You’ll end up tuning this value to get the behavior you desire. A small value can
make the rope seem jittery, while a large value will make the stretchiness appear
smoother. Higher damping also helps alleviate numerical instabilities to some extent,
although it’s no substitute for a robust integration scheme.
260 | Chapter 13: Connecting Objects
www.it-ebooks.info
Initialize the particles and springs
Initially, our particle rope is set up horizontally, as shown in Figure 13-3, with the left¬
most particle, pO, fixed—that is, the particle pO will not move, and the remainder of the
rope will pivot about pO. For convenience, all remaining particles are incrementally
indexed from left to right.
pO pi p2 p3 p4 p5 p6 p7 p8 p9
sO si s2 s3 s4 s5 s6 s7 s8
Figure 13-3. Particle rope setup
There are nine springs, which are indexed from left to right as illustrated in
Figure 13-3. Spring 0 connects particle 0 to particle 1, spring 1 connects particle 1 to
particle 2, and so on. The following code sample shows how all this is initialized:
boot Inttiallze(void)
1
Vector r;
int 1;
Objects[0].bLocked = true;
// Initialize particle locations from left to right.
for(i=0; i<_NUM_OBJECTS; U+)
{
Objectsfi].vPosition.x = _WINWIDTH/2 + Objects[0].fLength * i;
Objectsfi].vPosition.y = _WINHEIGHT/8;
}
// Initialize springs connecting particles from left to right.
for(i=0; i<_NUM_SPRINGS; i++)
{
Springsfi].Endl = i;
Springsfi].End2 = i+1;
r = Objects[i+l].vPosition - Objectsfi].vPosition;
Springsfi].InitialLength = r.MagnitudeQ;
Springs[i].k = _SPRING_K;
Springs[i].d = _SPRING_D;
}
Connecting Particles | 261
www.it-ebooks.info
return true;
}
First, the local variables r and i are declared, r will be used to compute the initial,
unstretched length of the springs, and i will be used to index the Objects and Springs
arrays. Second, Objects [0], the one that is fixed, has its bLocked property set to true,
indicating that it does not move (that is, it’s locked).
Next, the particle positions are initialized starting from the first particle—the fixed one
positioned at the middle of the screen—and proceeding to the rest of the particles,
offsetting each to the right by an amount equal to property fLength. fLength is an
arbitrary length that you can define, that represents the spacing of the particles and,
subsequently, the initial length of the springs connecting the particles.
Finally, we set up the springs, connecting each particle to its neighbor on the right.
Starting at the first spring, we set its end references to the index of the particle to its left
and right in the properties Endl and End2, respectively. These indices are simply i and
i+l, as shown in the preceding code sample within the last for loop. The initial length
vector of the spring is computed and stored in the vector r, where r = Objects [i
+1]. vPosition - Objectsfi]. vPosition. The magnitude of this vector is the initial
spring length, which is stored in the spring’s property, InitialLength. This step isn’t
strictly necessary in this example since you already know that the property fLength
discussed earlier is the initial length of each spring. However, we’ve done it this general
way since you may not necessarily initialize the particle positions as we have simply
done.
Update the simulation
Updating the particle positions at each simulation time step, under the influence of
gravity and spring forces proceeds just like in the earlier examples of Chapter 8 and
Chapter 9. Essentially, you must compute the forces on the particles, integrate the equa¬
tions of motion, and redraw the scene. As usual in our examples, the function Update
Simulation is called on to perform these tasks. For the current example, UpdateSimu
lation looks like this:
void UpdateSimuLation(void)
{
double
dt = _TIMESTEP;
int
i;
double
f, dl;
Vector
ptl, pt2;
int
j;
Vector
r;
Vector
F;
Vector
vl, v2, vr;
// Initialize the spring forces on each object to zero.
for(i=0; i<_NUM_OBJECTS; i++)
262 | Chapter 13: Connecting Objects
www.it-ebooks.info
{
Objects[i].vSprings.x = 0;
Objects[i].vSprings.y = 0;
Objectsfi].vSprings.z = 0;
}
// Calculate all spring forces based on positions of connected objects.
for(i=0; i<_NUM_SPRINGS; i++)
{
j = Springs[i].Endl;
ptl = Objectsfj].vPosition;
vl = Objects[j].vVelocity;
j = Springs[i].End2;
pt2 = Objects[j].vPosition;
v2 = Objectsfj].vVelocity;
vr = v2 - vl;
r = pt2 - ptl;
dl = r.Magnitude() - Springs[i].InitialLength;
f = Springs[i].k * dl; // - means compression, + means tension
r.Normalize();
F = (r*f) + (Springsfi].d*(vr*r))*r;
j = Springs[i].Endl;
Objects[j].vSprings += F;
j = Springs[i].End2;
Objects[j].vSprings -= F;
// Integrate equations of motion as usual.
// Render the scene as usual.
}
As you can see, there are several local variables here. We’ll explain each one as we get
to the code where it’s used. After the local variable declarations, this function’s first task
is to reset the aggregate spring forces on each particle to 0. Each particle stores the
aggregate spring force in the property vSprings, which is a vector. In this example, each
particle will have up to two springs acting on it at any given time.
Connecting Particles | 263
www.it-ebooks.info
The next block of code in the for loop computes the springs forces acting on each
particle. There are several steps to this, so we’ll go through each one. First the loop is
set up to step through the list of springs. Recall that each spring is connected to two
particles, so each step through the loop will compute a spring force and apply it to two
separate particles.
Within the loop, the variable j is used as a convenience to temporarily store the index
that refers to the Object to which the spring is attached. For each spring j is first set to
the spring’s Endl property. A temporary variable, ptl, is then set equal to the position
of the Object to which j refers. Another temporary variable, vl, is set to the velocity of
the Object to which j refers. Next, j is set to the index of End2, the other Object to
which the current spring is attached, and that object’s position and velocity are stored
in pt2 and v2, respectively. This sort of temporary variable use isn’t necessary, of course,
but it makes the following lines of code that compute the spring force more readable in
our opinion.
vr is a vector that stores the relative velocity between the two ends of the spring. We
compute vr by subtracting vl from v2. Similarly, r is a vector that stores the relative
distance between the two ends of the spring. We compute r by subtracting ptl from
pt2. The magnitude of r represents the stretched or compressed length of the spring.
The change in spring length is computed and stored in d l as follows:
dl = r.Magnitude!) - Springs[i].InittalLength;
dl will be negative if the computed length is shorter than the initial length of the spring.
This implies that the spring is in compression and should act to push the particles away
from each other. A positive dl means the spring is in tension and should act to pull the
particles toward each other. The line:
f = Springs[t].k * dl;
computes the corresponding spring force as a function of dl and the spring constant.
Note that f is a scalar and we have not yet computed its line of action, although we know
it acts along the line connecting the particles at Endl and End2. That line is represented
by r, which we computed earlier. And the spring force is just f times the unit vector
along r. Since we’re including damping, we have to use the spring-damper equation for
the total force acting on each particle, which we call the vector F. F is computed as
follows:
F = (r*f) + (Springs[i].d*(vr*r))*r;
The first term on the right side of the equals sign is the Hooke’s law-based spring force,
and the second term is the damping force. Note here that r is a unit vector previously
computed using the line:
r.Normalize!);
264 | Chapter 13: Connecting Objects
www.it-ebooks.info
Finally, the spring force is applied to each particle connected by the spring. Remember,
the force is equal in magnitude but opposite in direction for each particle. The lines:
j = Springs[t].Endl;
Objects!j].vSprings += F;
apply the spring force to the particle at the first end of the spring, whereas the lines:
j = Springs[i].End2;
Objects[j].vSprings -= F;
apply the opposite spring force to the particle at the second end of the spring.
That’s it for computing and applying the spring forces. The remainder of the code is
business as usual, where we compute the force due to gravity and add it to the aggregate
spring force for each particle and then integrate the equations of motion. Finally, we
render the scene at each time step.
Connecting Rigid Bodies
As with particles, you can connect rigid bodies with springs to simulate some interesting
things. For example, you may want to simulate something as simple as a linked chain,
where each link is connected to the other in series. Or perhaps you want to simulate
connected body parts to simulate rag doll physics or maybe a golfer’s swing. All these
require some means of connecting rigid bodies. In this section we’ll show you how to
use linear spring-dampers, the same we’ve discussed already, to connect rigid bodies.
We’ll start with a simple analog to the rope example discussed earlier. Instead of con¬
necting particles with springs to simulate a dangling rope, we’ll connect rigid links to
simulate a dangling rope or chain. Later, we’ll show you how linear springs can be used
to restrain angular motion.
Links
In this example, each link is rigid in that it does not deform; however, the links are
connected by springs in a way that allows the ensemble to swing, stretch, and bend in
a manner similar to a hanging chain. Figure 13-4 illustrates our swinging linked chain
as it swings from right to left and then back toward the right.
Connecting Rigid Bodies | 265
www.it-ebooks.info
Figure 13-4. Swinging links
As in the rope example, the topmost link is connected to a fixed point by a spring, such
that the linked chain pivots around and hangs from the fixed point. The rectangles
represent each rigid link, with the lines connecting the rectangles representing springs.
To model this linked chain, we need only make a few changes to the rope example to
address the fact that we’re now dealing with rigid bodies that can rotate versus particles.
This requires us to specify the point on each body to which the springs are attached,
and in addition to computing the spring forces acting on each body, we must also com¬
pute the moments due to those forces. Aside from these spring force and moment
computations, the remainder of the simulation is the same as those discussed in Chap¬
ter 7 through Chapter 12.
Basic structures and variables
We can use the Spring structure shown earlier in the rope example again here with one
small modification. Basically, we need to change the type of the endpoint references,
Endl and End2, from integers to a new structure we’ll call EndPoint. The new Spring
structure looks like this:
typedef struct _Spring {
EndPoint Endl;
EndPoint End2;
float k;
266 | Chapter 13: Connecting Objects
www.it-ebooks.info
float d;
float InitlalLength;
} Spring, *pSpring;
The new EndPolnt structure is as follows:
typedef struct _EndPolntRef {
int ref;
Vector pt;
} EndPoint;
Here, ref is the index referring to the Object to which the spring is attached, and pt is
the point in the attached Objects local coordinate system to which the spring is attached.
Notice from Figure 13-4 that the first spring, the topmost one, is connected to a single
object; the other end of it is connected to a fixed point in space. We’ll use a ref of-1 to
indicate that a spring’s endpoint is connected to a fixed point in space instead of an
object.
As in the rope example, we have a few important defines and variables to set up:
#define _NUM_OBJECTS 10
#define _NUM_SPRINGS 10
#deflne _SPRING_K 1000
#define _SPRING_D 100
RlgidBody2D Objects[_NUM_OBJECTS];
Spring Springs[_NUM_SPRINGS];
These are the same as before except now we have 10 springs instead of 9, and Objects
is of type RigidBody2D instead of Particle.
The damping and spring constants play the same role here as they did in the rope
example.
Initialize
Initially our linked chain is set up horizontally, just like the rope example, but with the
link and spring indices shown in Figure 13-5. Each rectangle represents a rigid link, and
a spring attached to the left end of each link connects the link to its neighbor to the left.
In the case of the first link, L0, the spring connects the left end of the link to a fixed point
in space.
Connecting Rigid Bodies | 267
www.it-ebooks.info
■This end attached to fixed point in space
i LO LI L2 L3 L4 L5 L6 L7 L8 L9
.. 1 1 1 1 1
sO si s2 s3 s4 s5 s6 s7 s8 s9
Figure 13-5. Linked-chain setup
The code for this setup is only a little more involved than that for the rope example; the
additional complexity is due to having to deal with specific points on the rigid bodies
to which each spring is attached. The following code sample contains the modified
Initialize function:
boot Initialize(void)
{
Vector r;
Vector pt;
int i;
// Initialize objects for linked chain.
for(i=0; i<_NUM_LINKS; i++)
{
Objectsfi].vPosition.x = _WINWIDTH/2 + Objects[0].fLength * i;
Objectsfi].vPosition.y = _WINHEIGHT/8;
Objectsfi].fOrientation = 0;
}
// Connect end of the first object to a fixed point in space.
Springs[0].Endl.ref = -1;
Springs[0].Endl.pt.x = _WINklIDTH/2-Objects[0]. fLength/2;
Springs[0].Endl.pt.y = JJINHEIGHT/8;
Springs[0].End2.ref = 0;
Springs[0].End2.pt.x = -Objects[0].fLength/2;
Springs[0].End2.pt.y = 0;
pt = VRotate2D(Objects[0].fOrientation, Springs[0].End2.pt)
+ Objects[0].vPosition;
r = pt - Springs[0].Endl.pt;
Springs[0]. InftialLength = r.MagnitudeQ;
268 | Chapter 13: Connecting Objects
www.it-ebooks.info
Sprlngs[0].k = _SPRING_K;
Springs[0].d = _SPRING_D;
// Connect end of all remaining springs.
for(i=l; 1<_NUM_LINKS; i++)
{
Sprlngsfl].Endl.ref = i-1;
Sprlngsfl].Endl.pt.x = Objects[i-l].fLength/2;
Sprlngsfl].Endl.pt.y = 0;
Springsfi].End2.ref = i;
Springsfi].End2.pt.x = -Objects[i].fLength/2;
Springsfi].End2.pt.y = 0;
pt = VRotate2D(Objects[i].fOrientation, Springs[i].End2.pt)
+ Objectsfl].vPosition;
r = pt - (VRotate2D(Objects[i-l].fOrlentatlon, Springs[i].Endl.pt)
+ Objects[i-l].vPosition);
Springsfi].InitialLength = r.MagnltudeQ;
Springs[i].k = _SPRING_K;
Springs[i].d = _SPRING_D;
}
return true;
}
The local variables r and I are the same as before; however, there’s a new variable, pt,
that we use to temporarily store the coordinates of specific points when converting from
one coordinate system to another. We’ll see how this is done shortly.
After the local variables are declared, the Object positions are initialized, starting from
the first Object positioned at the middle of the screen and proceeding to the rest of the
Objects, offsetting each to the right by an amount equal to property fLength. Here,
f Length is an arbitrary length representing the length of each rigid body, not the length
of the springs connecting each rigid body. As you’ll see momentarily, the initial length
of all the springs in this example is 0.
You should be aware that the coordinates for each object computed here are the coor¬
dinates of the object’s center of gravity, which in this example we defined as the middle
of the rectangle representing each object. Since these are rigid bodies, not only must
you specify their initial positions, but you must also specify their initial orientations as
shown in the preceding code sample. The way we have this example set up, each object
is initialized with an orientation of 0 degrees.
The next task is to set up the spring connecting the first link, the one on the left, to a
fixed point in space. The following code handles this task:
// Connect end of the first object to a fixed point In space.
Sprlngs[0].Endl.ref = -1;
Connecting Rigid Bodies | 269
www.it-ebooks.info
Sprlngs[0].Endl.pt.x = _WINWIDTH/2-Objects[0].fLength/2;
Springs[0].Endl.pt.y = JJINHEIGHT/8;
Sprlngs[0].End2.ref = 0;
Springs[0].End2.pt.x = -Objects[0].fLength/2;
Sprtngs[0].End2.pt.y = 0;
pt = VRotate2D(Objects[0].fOrientation, Springs[0].End2.pt)
+ Objects[0].vPositlon;
r = pt - Springs[0].Endl.pt;
Springs[0]. InttialLength = r.MagnitudeQ;
Springs[0].k = _SPRING_K;
Sprtngs[0].d = _SPRING_D;
The first spring, Spring[0], has its first endpoint, Endl, set to refer to -1, which, as
explained earlier, means that this end of the spring is connected to some fixed point in
space. The location of the point, stored in the Endl. pt property, must be specified in
global coordinates as shown previously.
Now the second end of the first spring is connected to the left end of the first link;
therefore, End2. ref of the first spring is set to 0, which is the index to the first Object.
The point on Object[0] to which the spring is attached is the leftmost end on the
centerline of the object; thus, its coordinates—relative to the object’s center of gravity
location and specified in local, body-fixed coordinates—are:
Springs[0].End2.pt.x = -Objects[0].fLength/2;
Sprtngs[0].End2.pt.y = 0;
Now remember, the points on Objects to which springs are attached are specified in
body-fixed, local coordinates of each referenced object, whereas any point fixed in space
to which a spring is attached and not on an Object must be specified in global, earth-
fixed coordinates. You have to keep these coordinates straight and make the appropriate
rotations when computing spring lengths throughout the simulation. The code;
pt = VRotate2D(Objects[0].fOrientation, Springs[0].End2.pt)
+ Objects[0].vPositlon;
r = pt - Springs[0].Endl.pt;
illustrates how to do this. To compute the initial spring length, we need to compute the
relative distance between the endpoints of the spring. In case of the first spring, Endl
was specified in global coordinates, but End 2 was specified in the local coordinate system
of Object [0]. Therefore, we have to convert the coordinates of End2 from local coor¬
dinates to global coordinates before calculating the relative distance between the ends.
The preceding line, which calls the VRotate2D function you saw in earlier chapters,
rotates the locally specified point, End2.pt, from local to global coordinates; it then adds
the Objects position to the result, arriving at a point, pt, in global coordinates coinci¬
dent with the second endpoint of the spring. The relative distance, r, is the second
endpoint, pt, minus the first endpoint, Endl. pt.
270 | Chapter 13: Connecting Objects
www.it-ebooks.info
Finally, we compute the initial length of the spring by taking the magnitude of r and
storing the result in the springs InitialLength property.
With the first spring out of the way, the Initialize function enters a loop to set up the
remaining springs. Proceeding from left to right in Figure 13-5, the first endpoint,
Endl, is connected to the right side of Object [i-1], and the second endpoint, End2, is
connected to the left side of Object [i]. Be aware that each endpoint of each spring is
specified in different coordinate systems. The left end is in the coordinate system of
Object[i-l], while the right end is in the coordinate system ofObject[i].It may seem
trivial during this setup, but when things start moving and rotating it is critically im¬
portant to keep these coordinate systems straight. Doing so involves transforming each
endpoint coordinate from the local system of the body to which it’s attached to the global
coordinate system. This is illustrated as follows:
pt = VRotate2D(0bjects[i].fOrientation, Springsfi].End2.pt)
+ Objectsfi].vPosition;
r = pt - (VRotate2D(0bjects[i-l].fOrientation, Springs[l].Endl.pt)
+ Objects[l-l].vPosition);
The first line converts the spring attachment point End2 from the local coordinate system
of Object[i] to global coordinates by performing a rotation and translation using
functions you’ve already seen numerous times now. The result is temporarily stored in
the local variable, pt. The second line converts the spring attachment point Endl from
the local coordinate system of Object [i-1] to global coordinates and then subtracts
the result from pt, yielding a vector, r, representing the relative distance between the
spring’s endpoints. The magnitude of r is the spring’s initial length. Performing these
same calculations during the simulation will result in the spring’s stretched or com¬
pressed length. That calculation is performed in UpdateSinulation.
Update
The function UpdateSimulation is substantially the same as that discussed in the rope
example. There are a few differences that we’ll highlight here. Again, these differences
are due to the fact that we’re now dealing with rigid bodies that rotate rather than simple
particles. The following code sample shows the additions to UpdateSimulation. You
can see there are a couple of new variables, M and Fo. M is used to temporarily store
moments due to spring forces Fo in the local coordinates of each Object.
Just as the property vSprings was initialized to 0 at the start of UpdateSimulation, so
too must we initialize vMSprings to 0. Recall, vSprings aggregates the spring forces
acting on each Object. For rigid bodies that rotate, we’ll use vMSprings to aggregate the
moments on each Object resulting from those spring forces:
void UpdateSinulatlon(void)
{
Connecting Rigid Bodies | 271
www.it-ebooks.info
Vector
Vector
M;
Fo;
// Initialize the spring forces and moments on each object to zero.
for(i=0; i<_NUM_OBJECTS; i++)
{
Objectsfi].vMSprings.x = 0;
Objectsfi].vMSprings.y = 0;
Objectsfi].vMSprings.z = 0;
// Calculate all spring forces based on positions of connected objects
for(i=0; i<_NUM_SPRINGS; i++)
{
if(Springs[i].Endl.ref == -1)
{
ptl = Springsfij.Endl.pt;
vl.x = vl.y = vl.z = 0; // point is not moving
} else {
j = Springs[i].Endl.ref;
ptl = Objects[j].vPosition + VRotate2D(0bjects[j].fOrientation,
Springs[i].Endl.pt);
vl = Objects[j].vVelocity + VRotate2D(0bjects[j].fOrientation,
Objects[j].vAngularVelocity A Springs[i].Endl.pt);
}
if(Springs[i].End2.ref == -1)
{
pt2 = Springsfi].End2.pt;
v2.x = v2.y = v2.z = 0;
} else {
j = Springs[i].End2.ref;
pt2 = Objects[j].vPosition + VRotate2D(0bjects[j].fOrientation,
Springs[i].End2.pt);
v2 = Objects[j].vVelocity + VRotate2D(0bjects[j].fOrientation,
Objects[j].vAngularVelocity A Springs[i].End2.pt);
}
// Compute spring-damper force,
vr = v2 - vl;
r = pt2 - ptl;
dl = r.Magnitude() - Springs[i].InitialLength;
f = Springs[i].k * dl;
r.Normalize();
F = (r*f) + (Springs[i].d*(vr*r))*r;
// Aggregate the spring force on each connected object
272 | Chapter 13: Connecting Objects
www.it-ebooks.info
j = Springs[i].Endl.ref;
if(j != -1)
Objects[j].vSprings += F;
j = Springs[i].End2.ref;
if(j != -1)
Objects[j].vSprings -= F;
// convert force to first ref local coords
// Get local lever
// calc moment
// Compute and aggregate moments due to spring force
// on each connected object,
j = Springs[i].Endl.ref;
lf(j != -1)
{
Fo = VRotate2D(-0bjects[j]. fomentation, F);
r = Springs[i].Endl.pt;
M = r A Fo;
Objects!j].vMSprlngs += M;
}
j = Springs[i].End2.ref;
if(j!= -1)
{
Fo = VRotate2D(-0bjects[j]. fomentation, F);
r = Springs[i].End2.pt;
M = r A Fo;
Objects[j].vMSprlngs -= M;
}
// Integrate equations of motion as usual.
// Render the scene as usual.
}
As in the rope example, UpdateSimulation steps through all the Springs, computing
their stretched or compressed length, the relative velocity of each spring’s endpoints,
and the resulting spring forces. These calculations are a bit different in this current
example because we have to handle rotation, as explained earlier.
Connecting Rigid Bodies | 273
www.it-ebooks.info
Upon entering the for loop in the preceding code sample, Endl of the current spring is
checked to see if it’s connected to a fixed point in space. If so, the temporary variable
ptl stores the global coordinates of the endpoint, and the variable vl stores the velocity
of the endpoint, which is 0. If the endpoint reference is a valid Object, then we compute
the position of the endpoint, stored in ptl, just like we did in the Initialize function,
using a coordinate transform as follows:
ptl = Objects!j].vPosition + VRotate2D(0bjects[j].fOrlentation,
Springs[i].Endl.pt);
We compute the velocity of that point as shown in Chapter 9 by first computing the
velocity of the point due to rotation in body-fixed coordinates, converting that to global
coordinates, and then adding the result to the Objects linear velocity. This is accom¬
plished in the following code:
vl = Objects!j].vVelocity + VRotate2D(0bjects[j].fOrlentation,
Objects!j].vAngularVelocity A Springs[i].Endl. pt);
We then repeat these calculations for the second endpoint of the spring.
Once we’ve obtained the positions and velocities of the spring endpoints, we compute
the spring-damper force in the same manner as in the rope example. The resulting spring
forces are aggregated in the vSprings property of each object. Note that if the spring
endpoint reference is a fixed point in space, we do not aggregate the force on that fixed
point.
Since the Objects are rigid bodies here, we now have to compute the moment due to
the spring force acting on each object. You must do this so that when we integrate the
equations of motion, the objects rotate properly.
For the Object connected to Endl of the current spring, the following lines compute
the moment:
Fo = VRotate2D(-0bjects[j].fOrlentation, F);
r = Springs[i].Endl.pt;
M = r A Fo;
Objects!j].vMSprings += M;
Fo is a vector representing the spring force computed earlier on the current Object in
the current Object’s local, body-fixed coordinate system. The line:
Fo = VRotate2D(-Objects!j].fOrlentation, F);
transforms F from global to local coordinates of the current Object, Object! j ].
r is set to the local, body-fixed coordinates of the spring attachment point for the current
Object, and we compute the resulting moment by taking the vector cross product of r
with Fo. The result is stored in the vector variable M, which gets aggregated in the Object
property vMSprings. We then perform these same sorts of calculations for the Object
connected to the other end of the spring.
274 | Chapter 13: Connecting Objects
www.it-ebooks.info
After these calculations, the rest of UpdateSimulation is the same as that shown earlier;
the function integrates the equations of motion and renders the scene.
Upon running this simulation, you’ll see the linked chain swing down and to the left
and then back and forth until the motion dampens out. You’ll also notice there’s some
stretch to the springs between the objects that appears to increase as you look from the
lower link to the upper link. This is indeed a non-uniform stretch in the springs, which
makes sense when you consider that the upper spring has more weight, thus more force,
pulling down on it than does the lower spring.
As in this rope example, you can tune the spring and damping constants to minimize
the spring stretch if that gap created by the stretched spring bothers you. You must keep
in mind numerical stability if your springs are too stiff, and here again, you must im¬
plement a robust integrator.
Rotational Restraint
So far we’ve used springs only to attach objects in a way that keeps the attachment points
together but allows the objects to rotate about the attachment point. This is a so-called
pinned joint. If you want a fixed joint that minimizes the amount of rotation between
the connected objects, you can add another spring to restrain the connected objects’
rotation.
Figure 13-6 illustrates an example comprising two rigid objects connected at their ends,
forming a ninety-degree angle. The uppermost end of the first object is connected to a
fixed point in space as in our rope and linked-chain examples. Under gravity, the as¬
sembly would rotate and swing around this fixed point. However, unlike the linked-
chain example, the extra spring prevents the lower link from pivoting around the other
end of the first link, as illustrated in Figure 13-7.
Figure 13-6. Rotation restraint setup
Connecting Rigid Bodies | 275
www.it-ebooks.info
Figure 13-7. Rotation restraint in action
The setup for this example is relatively straightforward and consists of setting the initial
positions and orientations of two rigid bodies and connecting three springs.
This example’s Initialize function is as follows:
boot Inttiallze(vold)
{
Vector r;
Vector pt;
int t;
// Position objects
Objects[0].vPosition.x = _WINWIDTH/2;
Objects[0].vPosition.y = _WINHEIGHT/8+Objects[0].fLength/2;
Objects[0].fOrlentation = 90;
Objects[l].vPosition.x = _WINWIDTH/2+0bjects[l].fLength/2;
Objects[l].vPosition.y = _WINHEIGHT/8+Objects[0].fLength;
Objects[l].fOrlentation = 0;
// Connect end of the first object to the earth:
Springs[0].Endl.ref = -1;
Springs[0].Endl.pt.x = _WINWIDTH/2;
Springs[0].Endl.pt.y = JJINHEIGHT/8;
Springs[0].End2.ref = 0;
Springs[0].End2.pt.x = -Objects[0].fLength/2;
Springs[0].End2.pt.y = 0;
pt = VRotate2D(Objects[0].fOrlentation, Springs[0].End2.pt) +
276 | Chapter 13: Connecting Objects
www.it-ebooks.info
Objects[0].vPosttion;
r = pt - Springs[0].Endl.pt;
Springs[0].InttialLength = r.MagnitudeQ;
Springs[0].k = _SPRING_K;
Springs[0].d = _SPRING_D;
// Connect other end of first object to end of second object
1 = 1;
Springs[i].Endl.ref = 1-1;
Springs[i].Endl.pt.x = Objects[i-l].fLength/2;
Springs[i].Endl.pt.y = 0;
Springs[i].End2.ref = i;
Springs[i].End2.pt.x = -Objects[i].fLength/2;
Springsfi].End2.pt .y = 0;
pt = VRotate2D(Objects[i].fOrientation, Springs[i].End2.pt) +
Objectsfi].vPosttion;
r = pt - (VRotate2D(0bjects[i-l].fOrientation, Springs[i].Endl.pt) +
Objects[i-l].vPosttion);
Springsfi].InttialLength = r.MagnitudeQ;
Springs[i].k = _SPRING_K;
Springs[i].d = _SPRING_D;
// Connect CG of objects to each other
Springs[2].Endl.ref = 0;
Springs[2].Endl.pt.x = 0;
Springs[2].Endl.pt.y = 0;
Springs[2].End2.ref = 1;
Springs[2].End2.pt.x = 0;
Springs[2].End2.pt.y = 0;
r = Objects[l].vPosition - Objects[0].vPosttion;
Springs[2]. InitialLength = r.MagnitudeQ;
Springs[2].k = _SPRING_K;
Springs[2].d = _SPRING_D;
}
The two Objects are positioned with the lines:
Objects[0].vPosttion.x = _WINWIDTH/2;
Objects[0].vPosttion.y = _WINHEIGHT/8+Objects[0].fLength/2;
Objects[0].fOrientation = 90;
Objects[l].vPosttion.x = _WINWIDTH/2+0bjects[l].fLength/2;
Objects[l].vPosttion.y = _WINHEIGHT/8+Objects[0].fLength;
Objects[l].fOrientation = 0;
Connecting Rigid Bodies | 277
www.it-ebooks.info
Basically, the first Object, Object[0], is located somewhere toward the top middle of
the screen with an initial rotation of ninety degrees so that it stands vertically. The second
Object, Object [1], is positioned so that it lies horizontally with its left end coincident
with the lower end of the first object. We’ll put a spring there momentarily, but first,
we’ll connect a spring to the upper end of the first object to connect it to a fixed point.
The following code takes care of that spring using the same techniques discussed earlier:
// Connect end of the first object to the earth:
Sprlngs[0].Endl.ref = -1;
Springs[0].Endl.pt.x = JJINWIDTH/2;
Sprlngs[0].Endl.pt.y = JJINHEIGHT/8;
Springs[0].End2.ref = 0;
Springs[0].End2.pt.x = -Objects[0].fLength/2;
Springs[0].End2.pt.y = 0;
pt = VRotate2D(Objects[0].fOrientation, Sprlngs[0].End2.pt) +
Objects[0].vPosttion;
r = pt - Sprlngs[0].Endl.pt;
Springs[0]. InltialLength = r.MagnitudeQ;
Springs[0].k = _SPRING_K;
Springs[0].d = _SPRING_D;
Now, we connect a spring at the corner formed by the two objects using the following
code:
// Connect other end of first object to end of second object
1 = 1 ;
Springs[i].Endl.ref = 1-1;
Springsfi].Endl.pt.x = Objects[l-l].fLength/2;
Springs[i].Endl.pt.y = 0;
Springs[i].End2.ref = i;
Springs[i].End2.pt.x = -Objectsfi].fLength/2;
Springs[i].End2.pt.y = 0;
pt = VRotate2D(0bjects[i].fOrientation, Springsfi].End2.pt) +
Objectsfi].vPosition;
r = pt - (VRotate2D(0bjects[i-l].fOrientation, Springsfi].Endl.pt) +
Objects[i-l].vPosition);
Springs[i]. InitialLength = r.MagnitudeQ;
Springs[i].k = _SPRING_K;
Springs[i].d = _SPRING_D;
If we stop here, the simulation will behave just like the linked-chain example, albeit we’ll
have a very short chain. So, to prevent rotation at the corner, we’ll add another spring
connecting the centers of gravity of the objects. You can use other points if you desire;
we chose the centers of gravity for convenience. The following code adds this rotational
restraint spring:
278 | Chapter 13: Connecting Objects
www.it-ebooks.info
// Connect CG of objects to each other
Springs[2].Endl.ref = 0;
Springs[2].Endl.pt.x = 0;
Sprlngs[2].Endl.pt.y = 0;
Sprtngs[2].End2.ref = 1;
Sprlngs[2].End2.pt.x = 0;
Springs[2].End2.pt.y = 0;
r = Objects[l].vPosition - Objects[0].vPosition;
Springs[2]. InttialLength = r.MagnitudeQ;
Springs[2].k = _SPRING_K;
Sprtngs[2].d = _SPRING_D;
The rest of this simulation is the same as in the linked-chain example. There are no
other code modifications required. It’s all in the setup.
Now, if you want to allow some amount of rotation or flexibility in the joint, you can
do so by tuning the spring constant for the rotation restraint spring. Using linear springs
creatively, you can model all sorts of joints very simply.
Connecting Rigid Bodies | 279
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 14
Physics Engines
A physics engine is the part of your game that contains all the code required for whatever
you’re trying to simulate using physics-based techniques. For many game programmers,
a physics engine is a real-time, rigid-body simulator such as the sort we’ve discussed
earlier in this book. The open source and licensable physics engines available to you are
typically of the rigid-body-simulator variety. Some physics engines are rather generic
and are useful for general rigid bodies and particles; others include various connectors
and constraints, enabling ragdoll simulation. Still others focus on soft bodies and fluids.
Fewer actually focus on the physics of some specific thing, like a car or a boat. A simple
Internet search on the phrase “game physics engine” will generate many links to potential
options for your use. That said, you could always write your own physics engine.
Building Your Own Physics Engine
We’re advocates of using physics where you need it. Sure, you can write a general-
purpose physics engine for a game, but if you’re creating a game that doesn’t require a
general-purpose physics engine, then don’t write one. That may sound obvious, but
sometimes we are compelled to do more than what we need just so we can say we did
it. Aside from the effort involved, a general-purpose physics engine will probably be
less efficient than a purpose-built physics engine. By purpose-built, we mean designing
the physics engine specifically to suit what you’re trying to simulate. For example, a
general-purpose physics engine would surely include particles, rigid bodies, connectors,
other force effectors, and who knows what else—fluids, perhaps—and be fully 3D. But
if you’re writing a 2D side-scrolling game for a smartphone, you certainly won’t need
3D with the associated complexities involved in dealing with rotation and collisions in
3D; and if your game simply involves throwing a ball of fuzz at some arbitrary junk,
then you may not even need to deal with rigid bodies at all. We’re being somewhat
facetious here, but the point is, unless you must write a general-purpose physics engine
—say, if you plan to license it as a middleware product or use it in a variety of game
281
www.it-ebooks.info
types—then don’t write one. Instead, write one specifically optimized for the game
you’re working on.
Let’s consider a few examples. Let’s say you’re writing a 3D first-person shooter and you
want to use physics to simulate how wooden barrels and crates blow apart when shot.
Typically, such an effect would show pieces of wood flying off in different directions
while falling under the influence of gravity. You could simulate such an effect in 3D
using rigid bodies and you wouldn’t even need to consider collisions, unless you wanted
the pieces to bounce off each other or other objects. Ignoring these aspects greatly sim¬
plifies the underlying physics engine. Consider another example. Let’s say you’re work¬
ing on a game involving flying an airplane. You can use physics to simulate the flight
dynamics, as we explain in this book, without the need for particles, connectors, or even
collision response.
The point of all this discussion is that you should consider which aspects of your game
will really benefit from physics and write your physics engine to deal specifically with
those aspects.
Another thing to consider is whether or not you need real-time physics. You might
expect, after reading the available game physics literature, that your game must include
real-time simulations if it is to incorporate physics. However, there are many ways to
include physics in a game without having to solve the physics via real-time simulations.
We show you an example in Chapter 19 whereby a golf swing is simulated in order to
determine club head velocity at the time of club-ball impact. In this case, given specific
initial parameters, we can solve the swing quickly, almost instantaneously, to determine
the club speed, which can then be used as an initial condition for the ball flight. The
ball’s flight can be solved quickly as well and not necessarily in real time. It really depends
on how you want to present the result to the player. If your game involves following the
flight of the ball as it soars through the air, then you might want to simulate its flight in
real time so you can realistically move the ball and camera. If, however, you simply want
to show where the ball ends up, then you need not perform the simulation in real time.
For such a simple problem, you can solve for the final ball location quicker than real
time. Sometimes, the action you’re simulating may happen so fast in real life that you’ll
want to slow it down for your game. Following a golf ball’s flight in real time might have
the camera moving so fast that your player won’t be able to enj oy the beautifully rendered
bird’s-eye view of the course. In this case, you rapidly solve the flight path, save the data,
and then animate the scene at a more enjoyable pace of your choosing.
We don’t want to come across as trying to talk you out of writing a physics engine if you
so choose. The point of our discussion so far is that you simply don’t have to write a
generic, real-time physics engine in order to use physics in your games. You have other
options as we’ve just explained.
Assuming that, after all these considerations, you need to write a physics engine, then
we have the following to offer.
282 | Chapter 14: Physics Engines
www.it-ebooks.info
A physics engine is just one component of a game engine. The other components include
the graphics engine, audio engine, AI engine, and whatever other engines you may
require or whatever other components of a game you may elevate to the status of engine.
Whatever the case, the physics engine handles the physics. Depending on whom you
talk to, you’ll get different ideas on what composes a physics engine. Some will say that
the heart of the physics engine is the collision detection module. Well, what if your game
doesn’t require collision detection, yet it still uses physics to simulate certain behaviors
or features? Then collision detection certainly cannot be the heart of your physics en¬
gine. Some programmers will certainly take issue with these statements. To them, a
physics engine simulates rigid-body motion using Newtonian dynamics while taking
care of collision detection and response. To us, a significant component of a physics
engine is the model—that is, the idealization of the thing you’re trying to simulate in a
realistic manner. You cannot realistically simulate the flight characteristics of a specific
aircraft by treating it as a generic rigid body. You have to develop a representative model
of that aircraft including very specific features; otherwise, it’s a hack (which, by the way,
we recognize as a valid and long-established approach).
Earlier, in Chapter 7 and Chapter 13, we showed you several example simulations. While
simple, these examples include many of the required components of a generic physics
engine. There are the particle and rigid-body classes that encapsulate generic object
properties and behaviors, physics models that govern object behaviors, collision detec¬
tion and response systems, and a numerical integrator. Additionally, those examples
include interfacing the physics code with user input and visual feedback. These examples
also show the basic flow from user input to physics solver to visual feedback.
In summary, the major components of a generic physics engine include:
• Physics models
• Simulated objects manager
• Collision detection engine or interface thereto
• Collision response module
• Force effectors
• Numerical integrator
• Game engine interface
Physics Models
Physics models are the idealizations of the things you’re simulating. If your physics
engine is a generic rigid-body simulator used to simulate an assortment of solid objects
your players can knock around, throw, shoot, and generally interact with in a basic
manner, then the physics model will probably be very generic. It’s probably safe to as-
Building Your Own Physics Engine | 283
www.it-ebooks.info
sume that each object will be subject to gravity’s pull, thus mass will be an important
attribute. Size will also be important, not only because you’ll need to know how big
things are when checking for collisions and handling other interactions, but also because
size is related to the distribution of the object’s mass. More precisely, each object will
have mass moment of inertia attributes. The objects will most likely also have some
ascribed coefficient of restitution that will be used during collision response handling.
Additionally, you might ascribe some friction coefficients that may be used during col¬
lision response or in situations where the objects may slide along a floor. As the objects
will likely find themselves airborne at some point, you’ll probably also include a drag
coefficient for each object. All of these parameters will help you differentiate massive
objects from lighter ones or compact objects from voluminous ones.
If your simulation involves more than generic rigid bodies, then your physics will be
more specific and perhaps far more elaborate. A great example of a more complicated
model is flight simulation. No matter how good your generic rigid-body model, it won’t
fly like any specific aircraft if it flies at all. You must develop a model that captures flight
aerodynamics specific to the aircraft you’re simulating. Chapter 15 shows how to put
together a model for an aircraft that can be used in a real-time flight simulation.
The other chapters in Part IV of this book are meant to give you a taste of modeling
aspects for a variety of things you might simulate in a game. Just as you cannot simulate
an aircraft with a generic rigid-body model, you cannot simulate a ship with an aircraft
model, nor can you simulate a golf ball with a ship model. The point is that you must
spend some time designing your physics model specific to what you’re going to simulate
in your game. Time spent here is just as important to creating a realistic physics engine
as time spent on designing a robust integration scheme or collision detection system.
We can’t overstate the importance of the physical model. The model is what defines the
behavior of the thing you’re simulating.
Simulated Objects Manager
Your simulated objects manager will be responsible for instantiating, initializing, and
disposing of objects. It will also be responsible for maintaining links between object
physics and other attributes such as geometry, for example, if in a 3D simulation you
use the same polyhedron to render an object and for collision detection and response.
You must have some means of managing the objects in your simulation. One can imag¬
ine many different approaches to managing these objects, and unless your simulation
uses just a handful of objects or fewer, essentially what you need is a list of objects of
whatever class you’ve defined. You’ve seen in previous chapters’ examples where we use
simple arrays of RigidBody type objects or Particle type objects. If all the objects in
your simulation are the same, then you need only a single class capturing all their be¬
havior. However, for more diversity, you should use a list of various classes with each
class encapsulating the code required to implement its own physical model. This is
284 | Chapter 14: Physics Engines
www.it-ebooks.info
particularly important with respect to the forces acting on the model. For example, you
could have some objects representing projectiles with others representing aircraft. These
different classes will share some common code (for example, collision detection); how¬
ever, the way forces are computed on each will vary due to the differences in how they
are modeled. With such an approach, each class must have code that implements its
particular model. During integration, the entire list will be traversed, calling the force
aggregation method for each object, and the particular class will handle the details suit¬
able for the type of object.
In some simple cases, you need not use different object classes if the types of objects
you plan to use are not too different. For example, it would be fairly straightforward to
implement a single class capable of handling both particles and rigid bodies. The object
class could include an obj ect type property used to denote whether the object is a particle
or a rigid body, and then the class methods would call the appropriate code. Again, this
will work satisfactorily for simple objects with few differences. If you want to simulate
more than two types of objects or if they are very different, you’re probably better off
using different classes specific to each object being simulated.
However you structure your classes or lists, the flow of processing your objects will
generally be the same. Every physics tick —that is, every time step in the physics simu¬
lation—you must check for object collisions, resolve those collisions, aggregate the usual
forces on each object, integrate the equations of motion for each object, and then update
each object’s state.
As we said, this is the general flow at every physics tick, or time step, which may not be
the same as your rendering steps. For example, for accuracy in your simulation you may
have to take small steps around a millisecond or so. You wouldn’t want to update the
graphics every millisecond when you need only about a third as many graphics updates
per second. Thus, your objects manager will have to be integrated with your overall
game engine, and your game engine must be responsible for making sure the physics
and graphics are updated appropriately.
Collision Detection
If collisions are an important part of your game, then a robust collision detection system
is required.
Your collision detection system is distinct from the collision response system or module,
though the two go hand in hand. Collision detection is the computational geometry
problem of determining if objects collide and, if so, what points are making contact.
These points are sometimes called the contact manifold. They’re just the points that are
touching, which could be a line or surface, though for simplicity usually the point, end
points, or points defining the contact surface boundary are all that are included in the
contact manifold.
Building Your Own Physics Engine | 285
www.it-ebooks.info
The collision detection system’s role is very specific: determine which objects are col¬
liding, what points on each object are involved in the collision, and the velocities of
those points. It sounds straightforward, but actual implementation can get quite com¬
plex. There are situations where fast-moving objects may go right though other objects,
especially thin ones, over a single time step, making the collision detection system miss
the collision if it relies solely on checking the separation distance between objects and
their relative velocity (i.e., it detects a collision if the objects are within some collision
tolerance and are also moving toward each other). A robust collision detection system
will capture this situation and respond accordingly. In Chapter 8 we simply check if
particles moved past the ground over the course of a single time step, for example, and
then reset their position to that of the ground plane level. We can handle many situations
using such simple techniques, especially when dealing with objects passing through
floors or walls; however, other situations may require more complex algorithms to pre¬
dict if a collision will occur sometime in the near future depending on how fast objects
are moving relative to each other. This latter case is called continuous collision detection,
and it is covered in many Internet, book, and technical paper sources. Many commercial
and open source physics engines advertise their capability to handle continuous colli¬
sion detection.
Another challenge associated with collision detection is the fact that it can be very time-
consuming if you have a large or even moderate number of objects in your simulation.
There are various techniques to deal with this. First, the game space is partitioned in
some coarse grid-like manner, and this grid is used to organize objects depending on
which cell they occupy. Then, in the second phase of collision detection, only those
objects occupying adjacent cells are checked against each other to see if they are col¬
liding. Without this grid partitioning, pairwise checks of every object against every other
object would be very computationally expensive. The second phase of collision detec¬
tion is often a broad approach using bounding spheres or bounding boxes, which may
be axis- or body-aligned. If the bounding spheres or boxes of each object are found to
collide, then the objects likewise may be colliding and further checks will be required;
otherwise, we can infer that the objects are not colliding. In the case of a potential
collision, these further checks become more complex depending on the geometry of the
objects. This phase generally involves polygon- and vertex-level checks; there are well-
established techniques for performing such checks that we won’t get into here. Again,
there’s a wealth of literature on collision detection available online.
Collision Response
Once the collision detection system does its job, it’s time for the collision response system
to deal with the colliding objects. Earlier in this book, we showed you how to implement
an impulse-moment collision response method. Recall that this method assumes that
at the instant of collision, the most significant forces acting on the objects are the col¬
lision forces, so all other forces can be ignored for that instant. The method then com-
286 | Chapter 14: Physics Engines
www.it-ebooks.info
putes the resulting velocities of the objects after colliding and instantly changes their
velocities accordingly. To perform the required calculations, the collision response sys¬
tem requires the objects colliding, of course, the collision points, and the velocities of
those points. Also, each object must have some associated mass and coefficient of res¬
titution, which are likewise used to compute the resulting velocities of the objects after
the collision.
In practice, the collision response system works hand in hand with the collision detec¬
tion system, particularly when dealing with objects that may be penetrating each other.
As we mentioned earlier, in many cases an object penetrating another, such as an object
penetrating a wall or floor, can simply be moved so that it is just touching the wall or
floor. Other cases maybe more complicated, and iterative algorithms are used to resolve
the penetration. For example, if penetration is detected, then the simulation may back
up to the previous time step and take a short time step to see if penetration still occurs.
If it does not, the simulation proceeds; otherwise, the simulation takes an even smaller
time step. This process repeats until penetration does not occur. This works fine in many
cases; however, sometimes the penetration is never resolved and the simulation could
get stuck taking smaller and smaller time steps. This failure to resolve could be attributed
to objects that are just outside the collision distance tolerance at one instant, and due
to numerical errors, exceedingly small time steps are required to stay outside of the
distance tolerance. Some programmers simply put an iteration limit in the code to
prevent the simulation from getting stuck, but the consequences in every situation may
be unpredictable. The continuous collision detection approach we mentioned earlier
avoids this sort of problem by predicting the future collision or penetration and dealing
with it ahead of time. Whatever the approach, there will be some back and forth and
data exchange between your collision detection and response systems to avoid excessive
penetration situations.
Additionally, there are situations when an object may come to rest in contact with an¬
other—for example, a box resting on the floor. There are many ways to deal with such
contact situations, one of which is to just allow the impulse-momentum approach to
deal with it. This works just fine in many cases; however, sometimes the objects in resting
contact will jitter with the impulse-momentum approach. One resolution to this jitter¬
ing problem is to put those objects to sleep—that is, if they are found to be colliding,
but their relative velocities are smaller than some tolerance, they are put to sleep. A
related but somewhat more complicated approach is to compute the contact normal
between the object and the floor and set that velocity to 0. This serves as a constraint,
preventing the object from penetrating the floor while still allowing it to slide along the
floor.
Force Effectors
Force effectors apply direct or indirect force on the objects in your simulations. Your
physics engine may include several. For example, if your engine allows users to move
Building Your Own Physics Engine | 287
www.it-ebooks.info
objects around with the mouse, then you’ll need some virtualization of the force applied
by the user via the mouse or a finger on a touch screen. This is an example of a direct
force. Another direct force effector could be a virtual jet engine. If you associate that
virtual engine, which produces some thrust force, with some object, then the associated
object will behave as though it were pushed around by the jet.
Some examples of indirect force effectors include gravity and wind. Gravity applies force
on objects by virtue of their mass, but it is typically modeled as body acceleration and
not an explicit force. Wind can be viewed as exerting a pressure force on an object, and
that force will be a function of the object’s size and drag coefficients.
You can imagine all sorts of force effectors, from ones similar to those just described to
perhaps some otherworldly ones. Whatever you imagine, you must remember that a
force has magnitude, direction, and some central point of application. If you put a jet
engine on the side of a box, the box will not only translate but will spin as well. Wind
creates a force that has a center of pressure, which is the point through which you can
assume the total wind force acts. The direction of the force and point of application are
important for capturing both translation and rotation. As an example, consider the
hovercraft we modeled in Chapter 9 that included two bow thrusters for steering and a
propeller for forward motion. Each of these direct force effectors—the bow thrusters
and the propeller—is applied at specific locations on the hovercraft. The bow thrusters
are located toward the bow and point sideways in order to create spin, thus allowing
some steering. The propeller is located on the center line of the hovercraft, which passes
through the hovercraft’s center of gravity so that it does not create spin and instead
simply pushes the craft forward. There’s another force effector in that model—aerody¬
namic drag, which is an indirect force effector. The drag force is applied at a point aft
of the center of gravity so that it creates some torque, or moment, which in this model
helps keep the hovercraft pointed straight; it provides some directional stability.
Whatever force effectors you contrive, they all must be aggregated for each object and
dealt with in your numerical integrator. Thus, your integrator must have some means
of accessing all the force effector information required to accurately simulate their effect
on each associated object.
Numerical Integrator
The integrator is responsible for solving the equations of motion for each object. We
showed you how to do this earlier, in Chapter 7 through Chapter 13. In your generic
physics engine you’ll iterate through all the objects in the simulation to compute their
new velocities, positions, and orientations at every time step. To make these calculations,
your integrator must have access to the force effectors associated with each object. The
forces will be used to compute accelerations, which will then be integrated to compute
velocities, and those in turn will be used to compute positions and orientations.
288 | Chapter 14: Physics Engines
www.it-ebooks.info
You can handle aggregating the forces in a few ways. You could aggregate all the forces
on all objects prior to looping through the objects integrating the equations of motion,
but this would require looping through all the objects twice. Alternatively, since you’re
looping through the object list in order to integrate each object’s equations of motion,
you can simply aggregate each object’s forces during the integration step. A complication
arises when object pairs apply forces to each other. A linear spring and damper, for
example, connected between two objects, apply equal and opposite forces to each object.
The force is a function of the relative distance between the objects (the spring compo¬
nent) and their relative velocity (the damping component). So, if during a given time
step you aggregate the forces on one of the objects in the pair, the resulting force will
be a function of the current relative position and velocities of the objects. Integrating
that object will give it a new position and velocity. Then, when you get to the other object
in the pair, if you recomputed the spring force, it will be a function of the new relative
position and velocity between the objects that includes the new displacement and ve¬
locity of the previously updated object but the old displacement and velocity of the
current object. This is inconsistent with Newton’s law of equal and opposite forces. You
can resolve this problem by storing the force computed for the first object and applying
it to the second one without recomputing the spring force for the second object.
Building Your Own Physics Engine | 289
www.it-ebooks.info
www.it-ebooks.info
PART III
Physical Modeling
Part III focuses on physical modeling. The aim of this part is to provide you with valuable
physical insight so you can make better judgments on what to include in your models
and what you can safely leave out without sacrificing physical realism. We cannot and
do not attempt to cover all the possible things you might want to simulate. Instead, we
cover several typical things you may try to simulate in a game—such as aircraft, boats,
and sports balls, among others—in order to give you some insight into their physical
nature and into some of the choices you must make when developing suitable models.
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 15
Aircraft
If you are going to write a flight simulation game, one of the most important aspects of
your game engine will be your flight model. Yes, your 3D graphics, user interface, story,
avionics system simulation, and coding are all important, but what really defines the
behavior of the aircraft that you are simulating is your flight model. Basically, this is
your simplified version of the physics of aircraft flight—that is, your assumptions, ap¬
proximations, and all the formulas you’ll use to calculate mass, inertia, and lift and drag
forces and moments.
There are four major forces that act on an airplane in flight: gravity, lift, thrust, and drag.
Gravity, of course, is the force that tends to pull the aircraft to the ground, while lift is
the force generated by the wings (or lifting surfaces) of the aircraft to counteract gravity
and enable the plane to stay aloft. The thrust force generated by the aircraft’s propulsor
(jet engine or propeller) increases the aircraft’s velocity and enables the lifting surfaces
to generate lift. Finally, drag counteracts the thrust force, tending to impede the aircraft’s
motion. Figure 15-1 illustrates these forces.
Lift
*
Gravity
Figure 15-1. Forces on aircraft inflight
293
www.it-ebooks.info
We’ve already discussed the force due to gravity in earlier chapters, so we won’t address
it again in this chapter except to say that the total of all lift forces must be greater than
or equal to the gravitational force if an aircraft is to maintain flight.
To address the other three forces acting on an aircraft, we’ll refer to a simplified, generic
model of an airplane and use it as an illustrative example. There are far too many aircraft
types and configurations to treat them all in this short chapter. Moreover, the subject
of aerodynamics is too broad and complex. Therefore, the model that we’ll look at will
be of a typical subsonic configuration, as shown in Figure 15-2.
Aileron
Straight, rectangular wing
Flap
Elevator
Propulsor Cockpit Fuselage
Span
Tail/Rudder
Figure 15-2. Model configuration
In this configuration the main lifting surfaces (the large wings) are located forward on
the aircraft, with relatively smaller lifting surfaces located toward the tail. This is the
basic arrangement of most aircraft in existence today.
We’ll have to make some assumptions in order to make even this simplified model
manageable. Further, we’ll rely on empirical data and formulas for the calculation of lift
and drag forces.
Geometry
Before getting into lift, drag, and thrust, we need to go over some basic geometry and
terms to make sure we are speaking the same language. Familiarity with these terms
294 | Chapter 15: Aircraft
www.it-ebooks.info
will also help you quickly find what you are looking for when searching through the
references that we’ll provide later.
First, take another look at the arrangement of our model aircraft in Figure 15-2. The
main body of the aircraft, the part usually occupied by cargo and people, is called the
fuselage. The wings are the large rectangular lifting surfaces protruding from the fuselage
near the forward end. The longer dimension of the wing is called its span, while its
shorter dimension is called its chord length, or simply chord. The ratio of span squared
to wing area is called the aspect ratio, and for rectangular wings this reduces to the ratio
of span-to-chord.
In our model, the ailerons are located on the outboard ends of the wings. The flaps are
also located on the wings inboard of the ailerons. The small wing-like surfaces located
near the tail are called elevators. And the vertical flap located on the aft end of the tail
is the rudder. We’ll talk more about what these control surfaces do later.
Taking a close look at a cross section of the wing helps to define a few more terms, as
shown in Figure 15-3.
The airfoil shown in Figure 15-3 is a typical cambered airfoil. Camber represents the
curvature of the airfoil. If you draw a straight line from the trailing edge to the leading
edge, you end up with what’s called the chord line. Now if you divide the airfoil into a
number of cross sections, like slices in a loaf of bread, going from the trailing edge to
the leading edge, and then draw a curved line passing through the midpoint of each
section’s thickness, you end up with the mean camber line. The maximum difference
between the mean camber line and the chord line is a measure of the camber of the
airfoil. The angle measured between the direction of travel of the airfoil (the relative
velocity vector of the airfoil as it passes through the air) and the chord line is called the
absolute angle of attack.
Geometry | 295
www.it-ebooks.info
When an aircraft is in flight, it may rotate about any axis. It is standard practice to always
refer to an aircraft’s rotations about three axes relative to the pilot. Thus, these axes—
the pitch axis, the roll axis, and the yaw axis—are fixed to the aircraft, so to speak,
irrespective of its actual orientation in three-dimensional space.
The pitch axis runs transversely across the aircraft—that is, in the port-starboard di¬
rection. 1 Pitch rotation is when the nose of the aircraft is raised or lowered from the
pilot’s perspective. The roll axis runs longitudinally through the center of the aircraft.
Roll motions (rotations) about this axis result in the wing tips being raised or lowered
on either side of the pilot. Finally, the yaw axis is a vertical axis about which the nose of
the aircraft rotates in the left-to-right (or right-to-left) direction with respect to the pilot.
These rotations are illustrated in Figure 15-4.
1. Port is to the pilots left and starboard is to the pilots right when he or she is sitting in the cockpit facing
forward.
296 | Chapter 15: Aircraft
www.it-ebooks.info
Lift and Drag
When an airfoil moves through a fluid such as air, lift is produced. The mechanisms by
which this occurs are similar to those in the case of the Magnus lift force, discussed
earlier in Chapter 6, in that Bernoulli’s law is still in effect. However, this time, instead
of rotation it’s the airfoil’s shape and angle of attack that affect the flow of air so as to
create lift.
Figure 15-5 shows an airfoil section moving through air at a speed V. V is the relative
velocity between the foil and the undisturbed air ahead of the foil. As the air hits and
moves around the foil, it splits at the forward stagnation point located near the foil
leading edge such that air flows both over and under the foil. The air that flows under
the foil gets deflected downward, while the air that flows over the foil speeds up as it
goes around the leading edge and over the surface of the foil. The air then flows smoothly
off the trailing edge; this is the so-called Kutta condition. Ideally, the boundary layer
remains “attached” to the foil without separating as in the case of the sphere discussed
in Chapter 6.
Relative air velocity
»
Velocity vector
M -
Low pressure.
High pressure . "'•CvC
Figure 15-5. Airfoil moving through air
The relatively fast-moving air above the foil results in a region of low pressure above
the foil (remember Bernoulli’s equation that shows pressure is inversely proportional
to velocity in fluid flow). The air hitting and moving along the underside of the foil
creates a region of relatively high pressure. The combined effect of this flow pattern is
to create regions of relatively low and high pressure above and below the airfoil. It’s this
pressure differential that gives rise to the lift force. By definition, the lift force is per¬
pendicular to the line of flight—that is, the velocity vector.
Lift and Drag | 297
www.it-ebooks.info
Note that the airfoil does not have to be cambered in order to generate lift; a flat plate
oriented at an angle of attack relative to the airflow will also generate lift. Likewise, an
airfoil does not have to have an angle of attack either. Cambered airfoils can generate
lift at 0, or even negative, angles of attack. Thus, in general, the total lift force on an
airfoil is composed of two components: the lift due to camber and the lift due to attack
angle.
Theoretically, the thickness of an airfoil does not contribute to lift. You can, after all,
have a thin curved wing as in the case of wings made from fabric (such as those used
for hang gliders). In practice, thickness is utilized for structural reasons. Further, thick¬
ness at the leading edge can help delay stall (more on this in a moment).
The pressure differential between the upper and lower surfaces of the airfoil also gives
rise to a drag force that acts in line with, but opposing, the velocity vector. The lift and
drag forces are perpendicular to each other and lie in the plane defined by the velocity
vector and the vector normal (perpendicular) to the airfoil chord line. When combined,
these two force components, lift and drag, yield the resultant force acting on the airfoil
in flight. This is illustrated in Figure 15-5.
Both lift and drag are functions of air density, speed, viscosity, surface area, aspect ratio,
and angle of attack. Traditionally, the lift and drag properties of a given foil design are
expressed in terms of nondimensional coefficients C L and C D , respectively:
C L = L / [(1/2) p V 2 S]
C D = D / [(1/2) p V 2 S]
where S is the wing planform area (span times chord for rectangular wings), L is the lift
force, D is the drag force, V is the speed through the air, and p (rho) is air density. These
coefficients are experimentally determined from wind tunnel tests of model airfoil de¬
signs at various angles of attack. The results of these tests are usually presented as graphs
of lift and drag coefficient versus attack angle. Figure 15-6 through Figure 15-8 illustrate
some typical lift and drag charts for a wing section.
298 | Chapter 15: Aircraft
www.it-ebooks.info
Lift and Drag | 299
www.it-ebooks.info
0.18
0.16
0.14
■S 0.12
.03
,§S
fj 0.1
o
CO
1 0.08
|
S 0.06
0.04
0.02
0
*■»
10 -5 0 5 10 15
Attack Angle (degrees)
C M
- - - - Cy with flap lowered 15 degrees
Figure 15-8. Typical C M versus attack angle
The most widely known family of foil section designs and test data is the NACA foil
sections. Theory of Wing Sections by Ira H. Abbott and Albert E. Von Doenhoff (Dover)
contains a wealth of lift and drag data for practical airfoil designs (see the Bibliogra¬
phy for a complete reference to this work). 2
In practice, the flow of air around a wing is not strictly two-dimensional—that is, flowing
uniformly over each parallel cross section of the wing—and there exists a span-wise
flow of air along the wing. The flow is said to be three-dimensional. The more three-
dimensional the flow, the less efficient the wing. 3 This effect is reduced on longer, high-
aspect-ratio wings (and wings with end plates where the effective aspect ratio is in¬
creased); thus, high-aspect-ratio wings are comparatively more efficient.
To account for the effect of aspect ratio, wing sections of various aspect ratios for a given
foil design are usually tested so as to produce a family of lift and drag curves versus
attack angle. There are other geometrical factors that affect the flow around wings; for
2. Theory of Wing Sections includes standard foil section geometry and performance data, including the well-
known NACA family of foil sections. The appendixes to Theory of Wing Sections have all the data you need
to collect lift and drag coefficient data for various airfoil designs, including those with flaps.
3. Lifting efficiency can be expressed in terms of lift-to-drag ratio. The higher the lift-to-drag ratio, the more
efficient the wing or foil section.
300 | Chapter 15: Aircraft
www.it-ebooks.info
a rigorous treatment of these, we refer you to the Theory of Wing Sections and Fluid-
Dynamic Lift. 4
Turning back to Figure 15-6, you’ll notice that the drag coefficient increases sharply
with attack angle. This is reasonable, as you would expect the wing to produce the most
drag when oriented flat against or perpendicular to the flow of air.
A look at the lift coefficient curve, which initially increases linearly with attack angle,
shows that at some attack angle the lift coefficient reaches a maximum value. This angle
is called the critical attack angle. For angles beyond the critical, the lift coefficient drops
off rapidly and the airfoil (or wing) will stall and cease to produce lift. This is bad. When
an aircraft stalls in the air, it will begin to drop rapidly until the pilot corrects the stall
situation by, for example, reducing pitch and increasing thrust. When stall occurs, the
air no longer flows smoothly over the trailing edge, and the corresponding high angle
of attack results in flow separation (as illustrated in Figure 15-9). This loss in lift is also
accompanied by an increase in drag.
Figure 15-9. Stalled airfoil
Theoretically, the resultant force acting on an airfoil acts through a point located at one-
quarter the chord length aft of the leading edge. This is called the quarter-chord point.
In reality, the resultant force line of action will vary depending on attack angle, pressure
distribution, and speed, among other factors. However, in practice it is reasonable to
assume that the line of action passes through the quarter-chord point for typical op¬
erational conditions. To account for the difference between the actual line of action of
the resultant and the quarter-chord point, we must consider the pitching moment about
the quarter-chord point. This pitching moment tends to tilt the leading edge of the foil
down. In some cases this moment is relatively small compared to the other moments
4. Fluid-Dynamic Lift, by Sighard F. Hoerner and Henry V. Borst, and Fluid-Dynamic Drag, by Sighard F.
Hoerner (both self-published by Hoerner), contain tons of practical charts, tables, and formulas for virtually
every aspect of aircraft aerodynamics. They even include material appropriate for high-speed boats and
automobiles.
Lift and Drag | 301
www.it-ebooks.info
acting on the aircraft, and it may be neglected. 5 An exception may be when the foil has
deflected flaps.
Flaps are control devices used to alter the shape of the foil so as to change its lift char¬
acteristics. Figure 15-6 also shows typical lift, drag, and moment coefficients for an
airfoil fitted with a plain flap deflected downward at 15°. 6 Notice the significant increase
in lift, drag, and pitch moment when the flap is deflected. Theory of Wing Sections also
provides data for flapped airfoils for flap angles between -15° and 60°.
Other Forces
The most notable force that we’ve yet to discuss is thrust—the propulsion force. Thrust
provides forward motion; without it, the aircraft’s wings can’t generate lift and the air¬
craft won’t fly. Thrust, whether generated by a propeller or a jet engine, is usually ex¬
pressed in pounds, and a common ratio used to compare the relative merits of aircraft
powering is the thrust-to-weight ratio. This ratio is the maximum thrust deliverable by
the propulsion plant divided by the aircraft’s total weight. When the thrust-to-weight
ratio is greater than one, the aircraft is capable of overcoming gravity in a vertical climb.
This is more like a rocket than a traditional airplane. Most normal planes are not capable
of this, but many military planes do have thrust-to-weight ratios of greater than one.
However, airplane engines rely on oxygen in the atmosphere to combust their fuel with
and to produce the force that propels them forward. As the plane climbs higher, the
engines will have less oxygen and produce less thrust. The thrust-to-weight ratio will
fall, and eventually the plane will again need lift from the wings to maintain its altitude.
Even when the plane is climbing vertically like a rocket, the wings still generate lift, and
in this case try to pull the airplane away from a vertical trajectory.
Besides gravity, thrust, wing lift, and wing drag, there are other forces that act on an
aircraft in flight. These are drag forces (and lift in some cases) on the various components
of the aircraft besides the wings. For example, the fuselage contributes to the overall
drag acting on the aircraft. Additionally, anything sticking out of the fuselage will con¬
tribute to the overall drag. If it’s not a wing, anything sticking out of the fuselage is
typically called an appendage. Some examples of appendages are the aircraft landing
gear, canopy, bombs, missiles, fuel pods, and air intakes.
Typically, drag data for fuselages and appendages is expressed in terms of a drag coef¬
ficient similar to that discussed in Chapter 6, where experimentally determined drag
forces are nondimensionalized by projected frontal area (S), density (p), and velocity
5. Aircraft designers must always consider this pitching moment when designing the aircrafts structure, as this
moment tends to want to twist the wings off the fuselage.
6. There’s a large variety of flap designs besides the plain trailing-edge flap discussed here. Flaps are typically
referred to in the literature as high lift devices , and the references we’ll provide in this chapter give rough
descriptions of the most common designs.
302 | Chapter 15: Aircraft
www.it-ebooks.info
squared (V 2 ). This means that the experimentally measured drag force is divided by the
quantity (1/2) p V 2 S to get the dimensionless drag coefficient. Depending on the object
under consideration, the drag coefficient data will be presented as a function of some
important geometric parameter, such as attack angle in the case of airfoils, or length-
to-height ratio in the case of canopies. Here again, Hoerner’s Fluid-Dynamic Drag is an
excellent source of practical data for all sorts of fuselage shapes and appendages.
For example, when an aircraft’s landing gear is down, the wheels (as well as associated
mechanical gear) contribute to the overall drag force on the aircraft. Hoerner reports
drag coefficients based on the frontal area of some small-plane landing-gear designs to
be in the range of 0.25 to 0.55. By comparison, drag coefficients for typical external
storage pods (such as for fuel), which are usually streamlined, can range from 0.06 to
0.26.
Another component of the total drag force acting on aircraft in flight is due to skin
friction. Aircraft wings, fuselages, and appendages are not completely smooth. Weld
seams, rivets, and even paint cause surface imperfections that increase frictional drag.
As in the case of the sphere data presented in Chapter 6. This frictional drag is dependent
on the nature of the flow around the part of the aircraft under consideration—that is,
whether the flow is laminar or turbulent. This implies that frictional drag coefficients
for specific surfaces will generally be a function of the Reynolds number.
In a rigorous analysis of a specific aircraft’s flight, you’d of course want to consider all
these additional drag components. If you’re interested in seeing the nitty-gritty details
of such calculations, we suggest you take a look at Chapter 14 of Fluid-Dynamic Drag,
where Hoerner gives a detailed example calculation of the total drag force on a fighter
aircraft.
Control
The flaps located on the inboard trailing edge of the wing in our model are used to alter
the chord and camber of the wing section to increase lift at a given speed. Flaps are used
primarily to increase lift during slow speed flight, such as when taking off or landing.
When landing, flaps are typically deployed at a high downward angle (downward flap
deflections are considered positive) on the order to 30° to 60°. This increases both the
lift and drag of the wings. During landing, this increase in drag also assists in slowing
the aircraft to a suitable landing speed. During takeoff, this increase in drag works
against you in that it necessitates higher thrust to get up to speed; thus, flaps may not
be deployed to as great an angle as when you are landing.
Ailerons control or induce roll motion by producing differential lift between the port
and starboard wing sections. The basic aileron is nothing more than a pair of trading-
edge flaps fitted to the tips of the wings. These flaps move opposite each other, one
deflecting upward and the other downward, to create a lift differential between the port
Control | 303
www.it-ebooks.info
and starboard wings. This lift differential, separated by the distance between the ailer¬
ons, creates a torque that rolls the aircraft. To roll the aircraft to the port side (the pilot’s
left), the starboard aileron would be deflected in a downward direction while the port
aileron would be deflected in an upward direction relative to the pilot. Likewise, the
opposite deflections of the ailerons would induce a roll to the starboard side. In a real
aircraft, the pilot controls the ailerons by moving the flight stick to either the left or
right.
Elevators, the tail “wings,” are used to control the pitch of the aircraft. (Elevators can be
flaps, as shown in Figure 15-2, or the entire tail wing can rotate as on the Lockheed
Martin F-16.) When the elevators are deflected such that their trailing edge goes down
with respect to the pilot, a nose-down pitch rotation will be induced; that is, the tail of
the aircraft will tend to rise relative to its nose, and the aircraft will dive. In an actual
aircraft, the pilot achieves this by pushing the flight stick forward. When elevators are
deflected such that their tailing edge goes up, a nose-up pitch rotation will be induced.
Elevators are very important for trimming (adjusting the pitch of) the aircraft. Generally,
the aircraft’s center of gravity is located above the mean quarter-chord line of the aircraft
wings such that the center of gravity is in line with the main lift force. However, as we
explained earlier, the lift force does not always pass through the quarter-chord point.
Further, the aircraft’s center of gravity may very well change during flight—for example,
as fuel is burned off and when ordnance is released. By controlling the elevators, the
pilot is able to adjust the attitude of the aircraft such that all of the forces balance and
the aircraft flies at the desired orientation (pitch angle).
Finally, the rudder is used to control yaw. The pilot uses foot pedals to control the rudder;
pushing the left (port) pedal yaws left and pushing the right pedals yaws right (star¬
board). The rudder is useful for fine-tuning the alignment of the aircraft for approach
on landing or when sighting up a target. Typically, large rudder action tends to also
induce roll motion that must be compensated for by proper use of the ailerons.
In some cases the rudder consists of a flap on the trailing edge of the vertical tail, while
in other cases there is no rudder flap and the entire vertical tail rotates. In both cases,
the vertical tail, which also provides directional stability, will usually have a symmetric
airfoil shape; that is, its mean camber line will be coincident with its chord line. When
the aircraft is flying straight and level, the tail will not generate lift since it is symmetric
and its attack angle will be 0. However, if the plane sideslips (yaws relative to its flight
direction), then the tail will be at an angle of attack and will generate lift, tending to
push the plane back to its original orientation.
304 | Chapter 15: Aircraft
www.it-ebooks.info
Modeling
Although we’ve yet to cover a lot of the material required to implement a real-time flight
simulator, we’d like to go ahead and outline some of the steps necessary to calculate the
lift and drag forces on your model aircraft:
1. Discretize the lifting surfaces into a number of smaller wing sections.
2. Collect geometric and foil performance data.
3. Calculate the relative air velocity over each wing section.
4. Calculate the attack angle for each wing section.
5. Determine the appropriate lift and drag coefficients and calculate lift and drag
forces.
The first step is relatively straightforward in that you need to divide the aircraft into
smaller sections where each section is approximately uniform in characteristics. Per¬
forming this step for the model shown in Figure 15-2, you might divide the wing into
four sections—one for each wing section that’s fitted with an aileron and one for each
section that’s fitted with a flap. You could also use two sections to model the elevators,
one port and one starboard, and another section to model the tail/rudder. Finally, you
could lump the entire fuselage together as one additional section or further subdivide
it into smaller sections depending on how detailed you want to get.
If you’re going to model your aircraft as a rigid body, you’ll have to account for all of
the forces and moments acting on the aircraft while it is in flight. Since the aircraft is
composed of a number of different components, each contributing to the total lift and
drag, you’ll have to break up your calculations into a number of smaller chunks and
then sum all contributions to get the resultant lift and drag forces. You can then use
these resultant forces along with thrust and gravity in the equations of motion for your
aircraft. You can, of course, refine your model further by adding more components for
such items as the cockpit canopy, landing gear, external fuel pods, bombs, etc. The level
of detail to which you go depends on the degree of accuracy you’re going for. If you are
trying to mimic the flight performance of a specific aircraft, then you need to sharpen
your pencil.
Once you’ve defined each section, you must now prepare the appropriate geometric and
performance data. For example, for the wings and other lifting surfaces you’ll need to
determine each section’s initial incidence angle (its fixed pitch or attack angle relative
to the aircraft reference system), span, chord length, aspect ratio, planform area, and
quarter-chord location relative to the aircraft’s center of gravity. You’ll also have to pre¬
pare a table of lift and drag coefficients versus attack angle appropriate for the section
under consideration. Since this data is usually presented in graphical form, you’ll have
to pull data from the charts to build your lookup table for use in your game. Finally,
Modeling | 305
www.it-ebooks.info
you’ll need to calculate the unit normal vector perpendicular to the plane of each wing
section. (You’ll need this later when calculating angle of attack.)
These first two steps need only be performed once at the beginning of your game or
simulation since the data will remain constant (unless your plane changes shape or its
center of gravity shifts during your simulation).
The third step involves calculating the relative velocity between the air and each com¬
ponent so you can calculate lift and drag forces. At first glance, this might seem trivial
since the aircraft will be traveling at an air speed that will be known to you during your
simulation. However, you must also remember that the aircraft is a rigid body and in
addition to the linear velocity of its center of gravity, you must account for its rotational
velocity.
Back in Chapter 2, we gave you a formula to calculate the relative velocity of any point
on a rigid body that was undergoing both linear and rotational motion:
v R = v cg + (o) X r)
This is the formula you’ll need to calculate the relative velocity at each component in
your model. In this case, v cg is the vector representing the air speed and flight direction
of the aircraft, o> (omega) is the angular velocity vector of the aircraft, and ris the distance
vector from the aircraft’s center of gravity to the component under consideration.
When dealing with wings, once you have the relative velocity vector, you can proceed
to calculate the attack angle for each wing section. The drag force vector will be parallel
to the relative velocity vector, while the lift force vector will be perpendicular to the
velocity vector. Angle of attack is then the angle between the lift force vector and the
normal vector perpendicular to the plane of the wing section. This involves taking the
dot product of these two vectors.
Once you have the attack angle, you can go to your coefficient of lift and drag versus
attack angle tables to determine the lift and drag coefficients to use at this instant in
your simulation. With these coefficients, you can use the following formulas to estimate
the magnitudes of lift and drag forces on the wing section under consideration:
Lift = C L (1/2) p V 2 S
Drag = C D (1/2) p V 2 S
This is a very simplified approach that only approximates the lift and drag character¬
istics. This approach does not account for span-wise flow effects, or the flow effects
between adjacent wing sections. Nor does this approach account for air disturbances,
such as downwash, that may affect the relative angle of attack for a wing section. Further,
the airflow over each wing section is assumed to be steady and uniform.
306 | Chapter 15: Aircraft
www.it-ebooks.info
As a simple example, consider wing panel 1, which is the starboard aileron wing section.
Assume that the wing is set at an initial incidence angle of 3.5° and that the plane is
traveling at a speed of 38.58 m/s in level flight at low altitude with a pitch angle of 4.5°.
This wing section has a chord length of 1.585 m and the span of this section is 1.829 m.
Using the lift and drag data presented in Figure 15-6, calculate the lift and drag on this
wing section, assuming the ailerons are not deflected and that the density of air is 1.221
kg/m 3 .
The first step is to calculate the angle of attack, which is 8°, based on the information
provided. Now, looking at Figure 15-6, you can find the airfoil lift and drag coefficients
to be 0.92 and 0.013, respectively.
Next, you’ll need to calculate the planform area of this section, which is simply its chord
times its span. This yields 2.899 m 2 . Now you have enough information to calculate lift
and drag as follows:
Lift = C L (1/2) p V 2 S
Lift = 0.92 (1/2) (1.221 kg/m 3 ) (38.58 m/s) 2 (2.899 m 2 )
Lift = 2,412.8 N
Drag = C D (1/2) p V 2 S
Drag = 0.013 (1/2) (1.221 kg/m 3 ) (38.58 m/s) 2 (2.899 m 2 )
Drag = 35.6 N
In your simulation, you’ll have to perform a similar set of calculations for every com¬
ponent that you’ve defined. As you can see, using this sort of empirical data and for¬
mulas, although only approximate, lends itself to fairly easy calculations. The hard part
is deciding what to model and finding the right data, and after that the lift and drag
calculations are pretty simple.
We’ve prepared an example program to show you how to model a simple airplane using
the method shown here. The program is named FlightSim.exe and is a real-time, 3D
flight simulator. The small aircraft simulated resembles that shown in Figure 15-2.
This program includes the following source files along with a text file ( Instructions.txt)
that explains the flight controls:
• Physics, cpp and Physics, h
• D3dstuff.cpp and D3dstuff.h
• Mymath.h
• Winmain.cpp
As we said, this program is a real-time simulation, and it treats the aircraft as a rigid
body. We’ve covered real-time simulations earlier in this book, and Chapter 12 in par-
Modeling | 307
www.it-ebooks.info
ticular covered some aspects of this flight simulation; therefore, some of the code to
follow will be familiar to you. In this present chapter, though, we’re going to focus on a
few specific functions that implement the flight model. These functions are contained
in the source file Physics.cpp.
The first function we want you to look at is CalcAirplaneMassProperties:
//-.//
// This model uses a set of eight discrete elements to represent the
// airplane. The elements are described below:
//
// Element 0
// Element 1
// Element 2
// Element 3
// Element 4
// Element 5
// Element 6
// Element 7
//
//
//
It
tt
//--
void
Outboard; port (left) wing section fitted with ailerons
Inboard; port wing section fitted with landing flaps
Inboard; starboard (right) wing section fitted with
landing flaps
Outboard; starboard wing section fitted with ailerons
Port elevator fitted with flap
Starboard elevator fitted with flap
Vertical tail/rudder (no flap; the whole thing rotates)
The fuselage
This function first sets up each element and then goes on to calculate
the combined weight, center of gravity, and inertia tensor for the plane.
Some other properties of each element are also calculated, which you'll
need when calculating the lift and drag forces on the plane.
.//
CalcAirplaneMassProperties(void)
float
mass;
Vector
vMoment;
Vector
CG;
int
t;
float
Ixx, Iyy, Izz, Ixy, Ixz, Iyz
float
in, di;
// Initialize the elements here
// Initially the coordinates of each element are referenced from
// a design coordinates system located at the very tail end of the plane,
// its baseline and center line. Later, these coordinates will be adjusted
// so that each element is referenced to the combined center of gravity of
// the airplane.
Element[0].fMass = 6.56f;
Element[0].vDCoords = Vector(14.5f,12.0f,2.5f);
Element[0].vLocallnertia = Vector(13.92f,10.50f,24.00f);
Element[0].flncidence = -3.5f;
Element[0].fDihedral = 0.0f;
Element[0].fArea = 31.2f;
Element[0].iFlap = 0;
Element[l].fMass = 7.31f;
Element[l].vDCoords = Vector(14.5f,5.5f,2.5f);
Element[l].vLocallnertia = Vector(21.95f,12.22f,33.67f);
308 | Chapter 15: Aircraft
www.it-ebooks.info
Elementfl].flncidence = -3.5f;
Element[l].fDihedral = 0.0f;
Elenentfl].fArea = 36.4f;
Elenentfl].iFlap = 0;
Element[2].fMass = 7.31f;
Elenent[2].vDCoords = Vector(14.5f,-5.5f,2.5f);
Elenent[2].vLocallnertia = Vector(21.95f,12.22f,33.67f);
Element[2].flncidence = -3.5f;
Element[2].fDihedral = 0.0f;
Elenent[2].fArea = 36.4f;
Element[2].iFlap = 0;
Element[3].fMass = 6.56f;
Elenent[3].vDCoords = Vector(14.5f,-12.0f,2.5f);
Elenent[3].vLocallnertia = Vector(13.92f,10.50f,24.00f);
Element[3].flncidence = -3.5f;
Element[3].fDihedral = 0.0f;
Elenent[3].fArea = 31.2f;
Elenent[3].iFlap = 0;
Element[4].fMass = 2.62f;
Elenent[4].vDCoords = Vector(3.03f,2.5f,3.0f);
Elenent[4].vLocallnertia = Vector(0.837f,0.385f,1.206f);
Element[4].flncidence = 0.0f;
Element[4].fDihedral = 0.0f;
Element[4].fArea = 10.8f;
Elenent[4].iFlap = 0;
Element[5].fMass = 2.62f;
Element[5].vDCoords = Vector(3.03f,-2.5f,3.0f);
Element[5].vLocallnertia = Vector(0.837f,0.385f,1.206f);
Element[5].flncidence = 0.0f;
Element[5].fDihedral = 0.0f;
Elenent[5].fArea = 10.8f;
Elenent[5].iFlap = 0;
Element[6].fMass = 2.93f;
Element[6].vDCoords = Vector(2.25f,0.0f,5.0f);
Element[6].vLocallnertia = Vector(1.262f,1.942f,0.718f);
Element[6].flncidence = 0.0f;
Elenent[6].fDihedral = 90.0f;
Elenent[6].fArea = 12.Of;
Element[6].iFlap = 0;
Element[7].fMass = 31.8f;
Element[7].vDCoords = Vector(15.25f,0.0f,1. 5f);
Element[7].vLocallnertia = Vector(66.30f,861.9f,861.9f);
Element[7].flncidence = 0.0f;
Element[7].fDihedral = 0.0f;
Elenent[7].fArea = 84.0f;
Elenent[7].iFlap = 0;
Modeling
| 309
www.it-ebooks.info
// Calculate the vector normal (perpendicular) to each lifting surface.
// This is required when you are calculating the relative air velocity for
// lift and drag calculations,
for (i = 0; i< 8; i++)
{
in = DegreesToRadians(Element[i].flncidence);
di = DegreesToRadians(Element[i].fDihedral);
Element[i].vNormal = Vector((float)sin(in), (float)(cos(in)*sin(di)),
(float)(cos(in)*cos(di)));
Element[i].vNormal.Normalize();
}
// Calculate total mass
mass = 0;
for (i = 0; i< 8; i++)
mass += Elementfi].fMass;
// Calculate combined center of gravity location
vMoment = Vector(0.0f, 0.0f, 0.0f);
for (i = 0; i< 8; i++)
{
vMoment += Element[i].fMass*Element[i].vDCoords;
}
CG = vMoment/mass;
// Calculate coordinates of each element with respect to the combined CG
for (i = 0; i< 8; i++)
{
Element[i].vCGCoords = Element[i].vDCoords - CG;
}
// Now calculate the moments and products of inertia for the
// combined elements.
// (This inertia matrix (tensor) is in body coordinates)
Ixx = 0; Iyy = 0; Izz = 0;
Ixy = 0; Ixz = 0; Iyz = 0;
for (i = 0; i< 8; i++)
{
Ixx += Element[i].vLocallnertia.x + Element[i].fMass *
(Element[i].vCGCoords.y*Element[i].vCGCoords.y +
Element[i].vCGCoords.z*Element[i].vCGCoords.z);
Iyy += Element[i].vLocallnertia.y + Element[i].fMass *
(Element[i].vCGCoords.z*Element[i].vCGCoords.z +
Element[i].vCGCoords.x*Element[i].vCGCoords.x);
Izz += Element[i].vLocallnertia.z + Element[i].fMass *
(Element[i].vCGCoords.x*Element[i].vCGCoords.x +
Element[i].vCGCoords.y*Element[i].vCGCoords.y);
Ixy += Element[i].fMass * (Element[i].vCGCoords.x *
Element[i].vCGCoords.y);
Ixz += Element[i].fMass * (Elementfi].vCGCoords.x *
Element[i].vCGCoords.z);
310 | Chapter 15: Aircraft
www.it-ebooks.info
Iyz += Element[i].fMass * (Element[i].vCGCoords.y *
Element[i].vCGCoords.z);
}
// Finally, set up the airplane's mass and its inertia matrix and take the
// inverse of the inertia matrix.
Airplane.fMass = mass;
Airplane.mlnertia.ell = Ixx;
Airplane.mlnertia.el2 = -Ixy;
Airplane.mlnertia.el3 = -Ixz;
Airplane.mlnertia.e21 = -Ixy;
Airplane.mlnertia.e22 = Iyy;
Airplane.mlnertia.e23 = -Iyz;
Airplane.mlnertia.e31 = -Ixz;
Airplane.mlnertia.e32 = -Iyz;
Airplane.mlnertia.e33 = Izz;
Airplane.mlnertialnverse = Airplane.mlnertia.Inverse();
}
Among other things, this function essentially completes step 1 (and part of step 2) of
our modeling method: discretize the airplane into a number of smaller pieces, each with
its own mass and lift and drag properties. For this model we chose to use eight pieces,
or elements, to describe the aircraft. Our comments at the beginning of the function
explain what each element represents.
The very first thing this function does is initialize the elements with the properties that
we’ve defined to approximate the aircraft. Each element is given a mass, a set of design
coordinates to its center of mass, a set of moments of inertia about each elements center
of mass, an initial incidence angle, a planform area, and a dihedral angle.
The design coordinates are the coordinates of the element with respect to an origin
located at the very tip of the aircraft’s tail, on its centerline and at its baseline. The x-axis
of this system points toward the nose of the aircraft, while the y-axis points toward the
port side. The z-axis points up. You have to set up your elements in this design coordinate
system first because you don’t yet know the location of the whole aircraft’s center of
mass, which is the combined center of mass of all of the elements. Ultimately, you want
each element referenced from the combined center of mass because it’s the center of
mass that you’ll track during the simulation.
The dihedral angle is the angle about the x-axis at which the element is initially set. For
our model, all of the elements have a 0 dihedral angle; that is, they are horizontal, except
for the tail rudder, which has a 90° dihedral since it is oriented vertically.
After we’ve set up the elements, the first calculation that this function performs is to
find the unit normal vector to each element’s surface based on the element’s incidence
and dihedral angles. You need this direction vector to help calculate the angle of attack
between the airflow and the element.
Modeling | 311
www.it-ebooks.info
The next calculation is the total mass calculation, which is simply the sum of all element
masses. Immediately following that, we determine the combined center of gravity lo¬
cation using the technique we discussed in Chapter 1. The coordinates to the combined
center of gravity are referenced to the design coordinate system. You need to subtract
this coordinate from the design coordinate of each element in order to determine each
element’s coordinates relative to the combined center of gravity. After that, you’re all set
with the exception of the combined moment of inertia tensor, which we already dis¬
cussed in Chapter 11 and Chapter 12.
Step 2 of our modeling method says you need to collect the airfoil performance data.
For the example program, we used a cambered airfoil with plain flaps to model the wings
and elevators, and we used a symmetric airfoil without flaps to model the tail rudder.
We didn’t use flaps for the tail rudder since we just made the whole thing rotate about
a vertical axis to provide rudder action.
For the wings, we set up two functions to handle the lift and drag coefficients:
//-.//
// Given the attack angle and the status of the flaps, this function
// returns the appropriate lift coefficient for a cambered airfoil with
// a plain trailing-edge flap (+/- 15 degree deflection).
//-.//
float LiftCoefficient(float angle, int flaps)
1
float clf0[9] = {-0.54f, -0.2f, 0.2f, 0.57f, 0.92f, 1.21f, 1.43f, 1.4f,
i.0f};
float clfd[9] = {0.0f, 0.45f, 0.85f, 1.02f, 1.39f, 1.65f, 1.75f, 1.38f,
1.17f};
float clfu[9] = {-0.74f, -0.4f, 0.0f, 0.27f, 0.63f, 0.92f, 1.03f, l.lf,
0.78f};
float a[9] = {-8.0f, -4.0f, 0.0f, 4.0f, 8.0f, 12.0f, 16.0f, 20.0f,
24.0f};
float cl;
int i;
cl = 0;
for (i=0; i<8; i++)
{
if( (a[i] <= angle) && (a[i+l] > angle) )
1
switch(flaps)
{
case 0:// flaps not deflected
cl = clf0[i] - (a[i] - angle) * (clf0[i] - clf0[i+l]) /
(a[i] - a[i+l]);
break;
case -1: // flaps down
cl = clfd[i] - (a[i] - angle) * (clfd[i] - clfd[i+l]) /
(a[i] - a[i+l]);
break;
312 | Chapter 15: Aircraft
www.it-ebooks.info
case 1: // flaps up
cl = clfu[l] - (a[1] - angle) * (clfu[l] - clfu[i+l]) /
(a[1] - a[i+l]);
break;
}
break;
}
}
return cl;
}
//.//
// Given the attack angle and the status of the flaps, this function
// returns the appropriate drag coefficient for a cambered airfoil with
// a plain trailing-edge flap (+/- 15 degree deflection).
//.//
float DragCoefficient(float angle, int flaps)
{
float cdf0[9]
float cdfd[9]
float cdfu[9]
float a[9]
float cd;
int i;
{0.01f, 0.0074f, 0.004f, 0.009f, 0.013f, 0.023f, 0.05f,
0.12f, 0.21f};
{0.0065f, 0.0043f, 0.0055f, 0.0153f, 0.0221f, 0.0391f, 0.1f,
0.195f, 0.3f};
{0.005f, 0.0043f, 0.0055f, 0.02601f, 0.03757f, 0.06647f,
0.13f, 0.18f, 0.25f};
= {-8.0f, -4.0f, 0.0f, 4.0f, 8.0f, 12.0f, 16.0f, 20.0f,
24.0f};
cd = 0.5;
for (i=0; i<8; i++)
{
if( (a[i] <= angle) && (a[i+l] > angle) )
{
switch(flaps)
{
case 0:// flaps not deflected
cd = cdf0[i] - (a[i] - angle) *
(a[i] - a[i+1]);
break;
case -1: // flaps down
cd = cdfd[i] - (a[i] - angle) *
(a[i] - a[i+1]);
break;
case 1: // flaps up
cd = cdfu[i] - (a[i] - angle) *
(a[i] - a[i+1]);
break;
}
break;
(cdf0[i] - cdf0[i+l]) /
(cdfd[i] - cdfd[i+l]) /
(cdfu[i] - cdfu[i+l]) /
Modeling | 313
www.it-ebooks.info
}
return cd;
}
Each of these functions takes the angle of attack as a parameter along with a flag used
to indicate the state of the flaps—that is, whether the flaps are in neutral position, de¬
flected downward, or deflected upward. Notice that the lift and drag coefficient data is
given for a set of discrete attack angles, thus we use linear interpolation to determine
the coefficients for attack angles that fall between the discrete angles.
The functions for determining the tail rudder lift and drag coefficients are similar to
those shown here for the wings, with the only differences being the coefficients them¬
selves and the fact that the tail rudder does not include flaps. Here are the functions:
//-.//
// Given the attack angle, this function returns the proper lift coefficient
// for a symmetric (no camber) airfoil without flaps.
//-.//
float RudderLiftCoefficient(float angle)
{
float clf0[7] = {0.16f, 0.456f, 0.736f, 0.968f, 1.144f, 1.12f, 0.8f};
float a[7] = {0.0f, 4.0f, 8.0f, 12.0f, 16.0f, 20.0f, 24.0f};
float cl;
int i;
float aa = (float) fabs(angle);
cl = 0;
for (i=0; i<8; i++)
{
if( (a[i] <= aa) && (a[i+1] > aa) )
{
cl = elf0[i] - (a[i] - aa) * (clf0[i] - clf0[i+l]) /
(a[i] - a[i+1]);
if (angle < 0) cl = -cl;
break;
}
}
return cl;
}
//-.//
// Given the attack angle, this function returns the proper drag coefficient
// for a symmetric (no camber) airfoil without flaps.
//-.//
float RudderDragCoefficient(float angle)
{
float cdf0[7] = {0.0032f, 0.0072f, 0.0104f, 0.0184f, 0.04f, 0.096f, 0.168f};
float a[7] = (0.0f, 4.0f, 8.0f, 12.0f, 16.0f, 20.0f, 24.0f};
float cd;
int i;
314 | Chapter 15: Aircraft
www.it-ebooks.info
float
aa = (float) fabs(angle);
cd = 0.5;
for (1=0; i<8; i++)
{
lf( (a[1] <= aa) && (a[i+l] > aa) )
{
cd = cdf0[l] - (a[i] - aa) * (cdf0[i] - cdf0[i+l]) /
(a[i] - a[i+l]);
break;
}
}
return cd;
}
With steps 1 and 2 out of the way, steps 3,4, and 5 are handled in a single function called
CalcAirplaneLoads:
//-.....//
// This function calculates all of the forces and moments acting on the
// plane at any given time.
//-.//
void CalcAirplaneLoads(void)
{
Vector Fb, Mb;
// reset forces and moments:
Airplane.vForces.x = 0.0f;
Airplane.vForces.y = 0.0f;
Airplane.vForces.z = 0.0f;
Airplane.vMoments.x = 0.0f;
Airplane.vMoments.y = 0.0f;
Airplane.vMoments.z = 0.0f;
Fb.x = 0.0f; Mb.x = 0.0f;
Fb.y = 0.0f; Mb.y = 0.0f;
Fb.z = 0.0f; Mb.z = 0.0f;
// Define the thrust vector, which acts through the plane's CG
Thrust.x = 1.0f;
Thrust.y = 0.0f;
Thrust.z = 0.0f;
Thrust *= ThrustForce;
// Calculate forces and moments in body space:
Vector
vLocalVelocity
float
fLocalSpeed;
Vector
vDragVector;
Vector
vLiftVector;
float
fAttackAngle;
float
tmp;
Vector
vResultant;
Modeling | 315
www.it-ebooks.info
int i;
Vector vtmp;
Stalling = false;
for(i=0; l<7; 1++) // loop through the seven lifting elements
// skipping the fuselage
{
if (i == 6) // The tail/rudder is a special case since it can rotate;
{ // thus, you have to recalculate the normal vector,
float in, di;
in = DegreesToRadians(Element[i].fIncidence); // incidence angle
di = DegreesToRadians(Element[i].fDihedral); // dihedral angle
Elementfi].vNormal = Vector( (float)sin(in),
(float)(cos(in)*sin(di)),
(float)(cos(in)*cos(di)));
Element[i].vNormal.Normalize();
}
// Calculate local velocity at element
// The local velocity includes the velocity due to linear
// motion of the airplane,
// plus the velocity at each element due to the
// rotation of the airplane.
// Here's the rotational part
vtmp = Airplane.vAngularVelocity A Element[i].vCGCoords;
vLocalVelocity = Airplane.vVelocityBody + vtmp;
// Calculate local air speed
fLocalSpeed = vLocalVelocity. MagnitudeQ;
// Find the direction in which drag will act.
// Drag always acts inline with the relative
// velocity but in the opposing direction
if(fLocalSpeed >1.)
vDragVector = -vLocalVelocity/fLocalSpeed;
// Find the direction in which lift will act.
// Lift is always perpendicular to the drag vector
vLiftVector = (vDragVector A Element[i].vNormal) A vDragVector;
tmp = vLiftVector. MagnitudeQ;
vLiftVector.Normalize();
// Find the angle of attack.
// The attack angle is the angle between the lift vector and the
// element normal vector. Note, the sine of the attack angle
//is equal to the cosine of the angle between the drag vector and
// the normal vector.
tmp = vDragVector*Element[i].vNormal;
if(tmp > 1.) tmp = 1;
316 | Chapter 15: Aircraft
www.it-ebooks.info
lf(tmp < -1) tmp = -1;
fAttackAngle = RadlansToDegrees((float) asin(tmp));
// Determine the resultant force (lift and drag) on the element,
tmp = 0.5f * rho * fLocalSpeed*fLocalSpeed * Element[i].fArea;
if (i == 6) // Tail/rudder
{
vResultant = (vLiftVector*RudderLiftCoefficient(fAttackAngle) +
vDragVector*RudderDragCoefficient(fAttackAngle))
* tmp;
} else
vResultant = (vLiftVector*LiftCoefficient(fAttackAngle,
Elementfi].iFlap) +
vDragVector*DragCoefficient(fAttackAngle,
Elementfi].iFlap) ) * tmp;
// Check for stall.
//We can easily determine stall by noting when the coefficient
//of lift is 0. In reality, stall warning devices give warnings well
// before the lift goes to 0 to give the pilot time to correct,
if (i<=0)
{
if (LiftCoefficient(fAttackAngle, Element[i].iFlap) == 0)
Stalling = true;
}
// Keep a running total of these resultant forces (total force)
Fb += vResultant;
// Calculate the moment about the CG of this element's force
// and keep a running total of these moments (total moment)
vtmp = Element[i].vCGCoords A vResultant;
lib += vtmp;
}
// Now add the thrust
Fb += Thrust;
// Convert forces from model space to earth space
Airplane.vForces = QVRotate(Airplane.qOrientation, Fb);
// Apply gravity (g is defined as -32.174 ft/s A 2)
Airplane.vForces.z += g * Airplane.fliass;
Airplane.vMoments += Mb;
}
The first thing this function does is reset the variables that hold the total force and
moment acting on the aircraft. Next, the thrust vector is set up. This is trivial in this
example since we’re assuming that the thrust vector always points in the plus x-axis
direction (toward the nose) and passes through the aircraft center of gravity (so it does
not create a moment).
Modeling | 317
www.it-ebooks.info
After calculating the thrust vector, the function loops over the model elements to cal¬
culate the lift and drag forces on each element. We’ve skipped the fuselage in this model;
however, if you want to account for its drag in your model, this is the place to add the
drag calculation.
Going into the loop, the first thing the function does is check to see if the current element
is element number six, the tail rudder. If it is, then the rudders normal vector is recal¬
culated based on the current incidence angle. The incidence angle for the rudder is
altered when you press the X or C keys to apply rudder action.
The next calculation is to determine the relative velocity between the air and the element
under consideration. As we stated earlier, this relative velocity consists of the linear
velocity as the airplane moves through the air plus the velocity of each element due to
the airplane’s rotation. Once you’ve obtained this vector, you calculate the relative air
speed by taking the magnitude of the relative velocity vector.
The next step is to determine the direction in which drag will act. Since drag opposes
motion, it acts inline with, but opposite to, the relative velocity vector; thus, all you need
to do is take the negative of the relative velocity vector and normalize the result (divide
it by its magnitude) to obtain the drag direction vector. Since this vector was normalized,
its length is equal to 1 (unity), so you can multiply it by the drag force that will be
calculated later to get the drag force vector.
After obtaining the drag direction vector, this function uses it to determine the lift
direction vector. The lift force vector is always perpendicular to the drag force vector,
so to calculate its direction you first take the cross product of the drag direction vector
with the element normal vector and then cross the result with the drag direction vector
again. Here again, the function normalizes the lift direction vector.
Now that the lift and drag direction vectors have been obtained, the function proceeds
to calculate the angle of attack for the current element. The attack angle is the angle
between the lift vector and the element normal. You can calculate the angle by taking
the inverse cosine of the vector dot product of the lift direction vector with the element
normal vector. Since the drag vector is perpendicular to the lift vector, you can get the
same result by taking the inverse sine of the vector dot product of the drag direction
vector with the element normal vector.
Now with all the lift and drag vector stuff out of the way, the function goes on to calculate
the resultant force acting on the element. The resultant force vector is simply the vector
sum of the lift and drag force vectors. Notice that this is where the lift and drag coefficient
functions are called and where the empirical lift and drag formulas previously discussed
are applied.
After calculating the resultant force, the function checks to see if the calculated lift
coefficient is 0. If it is, then the stall flag is set to warn us that the plane is in a stalled
situation.
318 | Chapter 15: Aircraft
www.it-ebooks.info
Finally, the resultant force is accumulated in the total force vector variable, and we
calculate the moment by taking the cross product of the element coordinate vector with
the resultant force. The resulting moment is accumulated in the total moment vector
variable. After exiting the loop, the function adds the thrust vector to the total force.
So far, all of these forces and moments have been referenced in the body-fixed-
coordinate system. The only thing left to do now is apply the gravity force, but this force
acts in the negative y-axis direction in the earth-fixed-coordinate system. To apply the
gravity force, the function must first rotate the body force vector from body space to
earth space coordinates. We used a quaternion rotation technique in this example, which
we already discussed in Chapter 11 and Chapter 12.
That’s pretty much it for the flight model. We encourage you to play with the flight model
in this program. Go ahead and tweak the element properties and watch to see what
happens. Even though this is a rough model, the flight results look quite realistic.
Modeling | 319
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 16
Ships and Boats
The physics of ships is a vast subject. While the same principles govern canoes and super
tankers, the difference between the two scales is not trivial. Our goal in this chapter will
be to explain some of the fundamental physical principles to allow you to develop re¬
alistic simulations. The typical displacement-type ship lends itself well to illustrating
these principles; however, many of these principles also apply to other objects sub¬
merged or partially submerged in a fluid, such as submarines and air balloons. Re¬
member, air is considered a fluid when we are considering buoyancy.
While surface ships or ships that operate on the water’s surface (at the air water interface)
are similar to fully submerged objects like submarines or air balloons in that they all
experience buoyancy, there are some very distinct differences in their physical nature
that we’ll highlight in this chapter. These differences affect their behavior, so it is im¬
portant to be aware of them if you intend to simulate such objects.
Ships have an entire language of their own, so we’ll be spending a lot of time just getting
the vocabulary right. This will allow you to do further research on any topics that are
of particular interest. There are many ways to classify ships and boats, but in regards to
the physics governing them, there are three basic types. Displacement vessels, semi¬
displacement vessels, and planing vessels are named after the forces that keep the boat
afloat while it is at cruising speed. When not moving, all vessels are in displacement
mode.
The term displacement in this context means that the ship is supported solely by buoy¬
ancy—that is, without dynamic or aerostatic lift as you would see on a high-speed racing
boat or a hovercraft. The word displacement itself refers to the volume of water displaced
or “pushed” out of the way by the ship as it sits floating in the water. We’ll discuss this
more in the next section.
A planing vessel is one that is not supported by buoyancy, but by hydrodynamic lift.
This includes the everyday speedboats that most boaters own. When the boat isn’t mov-
321
www.it-ebooks.info
ing, it just floats in the water, bobbing up and down. However, when the boat begins
traveling at high speed, the force of the water hitting the bottom of the boat causes the
boat to rise up. This is known as planing, and it greatly reduces the resistance of the
vessel. Semi-displacement vessels are those that straddle the two categories, with some
support coming from buoyancy and some coming from planing forces. Before we con¬
tinue discussing this, let’s go over some vocabulary.
The hull of the ship is the watertight part of the ship that actually displaces the water.
Everything in or on the ship is contained within the hull, which is partially submerged
in the water. The length of the ship is the distance measured from the bow to the stern.
In practice, there are several lengths used to denote the length of a ship, but here we’ll
refer to the overall length of the hull. The bow is the front of the ship, while the stern is
the aft part. When you are on the ship facing the bow, the port side is to your left and
the starboard side is to your right. The overall height of the hull is called the depth, and
its width is called breadth or beam. When a ship is floating in the water, the distance
from the water surface to the bottom of the hull is called the draft. Figure 16-1 illustrates
these terms.
Stern
Depth ~t
Draft
I
Beam
Bow
Hull
Profile
Port
Starboard
— Length—
Plan
Figure 16-1. Ship geometry
Given that ship design is a diverse subject, we’ll limit ourselves to discussing those
aspects of ships that make for realistic models. These subjects include stability and
sinking, resistance characteristics, propulstion, and manuverability. Most of these sub¬
jects cannot be fully simulated in real time, so we’ll show you some general rules that
ships follow instead of full numerical simulation.
322 | Chapter 16: Ships and Boats
www.it-ebooks.info
Stability and Sinking
If you have boats in your video game, the first step to making them realistic physically
is allowing them to sink if they become damaged. To understand why boats sink and
how they do so, you must first understand stability.
Stability
Most boats are least stable about their longitudinal axis—that is, they are easier to heel
port and starboard than they are to flip end over end. If the vessel heels over so far that
it is upside down, this is called capsizing. This is how most boats sink due to wind, waves,
or in some cases of side damage. One of the most famous examples of a sinking ship,
the Titanic, shows that when a boat is sinking from damage, it can sink end over end,
sometimes with the ship breaking in two. We’ll discuss both here so that you can animate
realistic sinking in your simulation.
In Chapter 3 we introduced the concept of buoyancy and stated that the force on a
submerged object due to buoyancy is a function of the submerged volume of the object.
Archimedes’s principle states that the weight of an object floating in a fluid is equal to
the weight of the volume of fluid displaced by the object. This is an important principle.
It says that a ship of a given weight must have sufficient volume to displace enough
water, an amount equal to the weight of the ship, in order for it to float. Further, this
principle provides a clever way of determining the weight of a ship: simply measure or
calculate the amount of water displaced by the ship and you can calculate the weight of
the ship. In the marine field, displacement is synonymous with the weight of the ship.
As discussed in Chapter 3, we can calculate the buoyant force on any object by using
the following formula:
F B = p g V
Here, V is the submerged volume of the object, p is the density of the fluid within which
the object is submerged, and g is the acceleration due to gravity. Since buoyancy is a
force, it has both magnitude and direction, and always acts straight up through the
center of buoyancy. The center of buoyancy is the geometric center of the submerged
part of the object.
When a ship is floating in equilibrium on the surface of the water, its center of buoyancy
must be located directly below the ship’s center of gravity. The weight of the ship, a force,
acts straight down through the center of gravity, opposing the force due to buoyancy.
When the ship is in equilibrium, these two forces, weight and buoyancy, are equal in
magnitude and opposite in direction.
Now, when an external force causes the ship to roll or pitch, the portion of the hull below
the water is changed and the center of buoyancy moves to the new geometric centroid
Stability and Sinking | 323
www.it-ebooks.info
of the underwater portion of the hull. For example, if the ship rolls to the starboard side,
then the center of buoyancy shiwfts out toward the starboard side. When this happens,
the lines of action of the weight of the ship and the buoyant force are no longer in line,
which results in a moment (torque) that acts on the ship. This torque is equal to the
perpendicular distance between the lines of action of the forces times the weight of the
ship.
Now here’s where we get to the floating upright part that we mentioned earlier. When
a ship rolls, for example, you don’t want it to keep rolling until it capsizes. Instead, you
want it to gently return itself to the upright position after whatever force caused it to
roll—the wind, for example—has been removed. In short, you want the ship to be stable.
For a ship to be stable, the line of action of the buoyant force must cross the vessel’s
centerline at a point, called the metacenter, above the center of gravity. When this hap¬
pens, the moment developed when the ship rolls tends to restore the ship to the upright
position. If the metacenter is located below the center of gravity, then the moment
developed would tend to capsize the ship. The distance between the center of gravity
and the metacenter is called GM. This is also known as the stability index, as a positive
value means the floating body is stable and a negative GM means the body is unstable.
Figure 16-2 illustrates these two scenarios.
Figure 16-2. Ship stability
324 | Chapter 16: Ships and Boats
www.it-ebooks.info
If you’re a sailor, then you know how important it is to keep the center of gravity of your
boat low. This helps increase the height of the metacenter above the center of gravity,
and thus helps with stability.
In the case of fully submerged objects, like submarines, the situation is different. The
buoyant force still acts through the geometric centroid of the object, but for stability,
the center of buoyancy must be located above the center of gravity. This way, when the
object rotates, the lines of action of the weight of the object and the buoyant force are
separated and form a moment that tends to restore the object to its upright position. If
it’s the other way around, then the object would be unstable, like trying to balance one
bowling ball on top of another. In this case, the slightest disturbance would upset the
balance and the object would flip upside-down such that the center of gravity is located
below the center of buoyancy.
Sinking
In general, boats protect their stability by compartmentalizing the hull into several wa¬
tertight sections, fittingly called compartments. This way, if the side of a vessel hits an
iceberg, only the compartment damaged will flood with seawater. If enough compart¬
ments are damaged, the vessel will not have enough buoyancy to support its weight and
it will sink. The end with the flooded compartments will sink first, causing a large angle
about the transverse axis. This is what happened to the Titanic. In fact, in that ship’s
case, the angle, called trim, was so large that the stern was lifted out of the water. The
hull could not support the weight of the stern section that was no longer being supported
by buoyancy, and the structure ripped in two.
It should be noted that ships can sink in the matter of minutes, or it can take hours. For
instance, the Titanic took about three hours to sink. The Lusitania sank in 18 minutes.
The time it takes depends heavily on the type of damage and the construction of the
vessel. We don’t suggest trying to get players to wait three hours for their game to end;
however, it is possible to continue fighting/propelling a vessel that is terminally dam¬
aged. In many cases where terminal damage is suspected, captains endeavor to ground
their vessels to prevent the ship from actually going under.
If side damage occurs, especially in high wind and waves, then it could be that the vessel
can still have enough buoyancy to float, but no longer enough stability to remain upright.
As damage usually occurs only on one side of a vessel, the center of buoyancy will no
longer be on centerline. This means that the restoring moment in one direction is di¬
minished by whatever amount the center has moved to that side. A big wave comes
along and pushes the vessel over to the point where the righting arm is no longer positive.
The vessel will flip 180 degrees with the bottom pointed skyward but will still float
(capsizing). Once rolled over, the remaining compartments will tend to fill with water
as vents or other openings fail over time. In the case of recreational boats, they are usually
Stability and Sinking | 325
www.it-ebooks.info
only a single compartment. If they capsize, they will sink readily; indeed, this is the way
that most small boats sink.
As we mentioned before, accurately computing all degrees of freedom for a nontrivial¬
shaped body in real time would be difficult to accomplish with todays computer hard¬
ware. In general, you want to follow a few high-level rules:
• The higher the center of gravity, the more likely it is that the boat will tip over.
• Large vessels are always compartmentalized. Damage should be limited to the wa¬
tertight compartment in which it occurred.
• The vessel will heal or trim in the direction of damage. If damage occurs on the
starboard side, the boat will heel to starboard. If the damage occurs in the bow, the
boat will list forward.
• A boat will remain floating as long as the undamaged compartments have a volume
in cubic meters of at least the weight of the hull in metric tons divided by 1.025.
• After being damaged, even if a vessel has enough undamaged volume to remain
afloat, it doesn’t necessarily mean it will float upright.
• Sinking almost never occurs as quickly as depicted in video games; however, cap¬
sizing can occur rapidly and is probably a more realistic way to model a stability
failure.
Ship Motions
Closely related to ship stability is the subject of ship motions. Knowing how vessels
work in a random set of waves will greatly help you to increase realism in your games.
The most important aspect of this is coupled motions, which we will talk about shortly.
First, some more vocabulary! As discussed before, there are six degrees of motion any
rigid body is capable of; for boats, some of these have special names and are described
next and illustrated in Figure 16-3.
326 | Chapter 16: Ships and Boats
www.it-ebooks.info
Roll, pitch, and yaw are the terms also used for airplanes. The translation degrees of
freedom are called surge, heave, and sway. Surge, sway, and yaw are not that apparent
when vessels are moving forward, so it is acceptable to limit your model to heave, pitch,
and roll. Heave is the up-and-down motion of the boat caused by the change in elevation
of the water’s surface as a wave passes. If a vessel is stationary, it would be referred to as
bobbing. Pitch is the rotation about the transverse axis of the vessel due to increased
buoyancy on one end of the ship as a wave passes. This motion is most pronounced
when the waves are traveling in the same direction (or 180 degrees) from the vessel. Roll
is like pitch, but about the longitudinal axis.
Heave
As stated before, heave is displacement in the vertical direction from the static equili¬
brium draft. This degree of freedom is straightforward to model as a hydrostatic spring
acting in the vertical direction. Assuming we have a barge that is 30 meters long and 10
meters wide, we’ll develop an equation that can govern our heave simulation.
Commonly, a vessel’s hydrostatics include something called tons per centimeter im¬
mersion (TPCM)—that is, for every centimeter you press the boat down, a certain
number of tons of buoyancy force is created. For our barge, this is a relatively straight¬
forward calculation.
Given that the water plane area is a constant 300 square meters, 1 centimeter of im¬
mersion would result in a volume of 3 cubic meters. As 1 cubic meter of saltwater weighs
1,027 kg, 3 cubic meters would be 3,081 kg, and (assuming this boat is on Earth), would
result in a buoyant force of 3,081 kg x 9.81 m/s 2 , or 30.2 kN. Therefore, 30.2 kN per cm
would make a good starting value for a spring constant to model the heave response of
this vessel in waves.
Roll
For us to simulate realistic roll motions, it is important that the ship take time to com¬
plete the motion. This time is called the roll period. This defines the angular velocity
that a ship rolled to one side will experience when it recovers. We can estimate it by the
following equation:
T =
2 nk
■s/gxGM
Where k is the radius of gyration and GM is the distance from the metacenter to the
center of gravity. A good estimate for the radius of gyration is often taken as 30% of the
beam of the vessel. A vessel with a shorter roll period will respond quicker to a wave
and try to assume the wave slope. This is known as being “stiff” and can cause passenger
discomfort and damage via higher angular accelerations. Conversely, vessels with higher
Ship Motions | 327
www.it-ebooks.info
roll periods are known as “tender” and lag behind the waves. These vessels generally
heel farther over but are more comfortable for passengers.
Pitch
Likewise, there is a pitch period that measures the speed at which the vessel responds
to a wave. This is highly dependent on the length of a ship and can be estimated as
follows:
T = (2it / g 1/2 )(k/(GM) 1/2 )
Where k is the radius of gyration and is commonly taken as 30% of the beam. Note that
pitch period is normally one-third to one-half that of roll period, so that the bow of a
vessel will rise and fall with the wave more in phase than in a roll event.
Coupled Motions
The real trick to getting motions to look right is to understand that for most boats the
motions are coupled. For instance, heave and pitch are strongly coupled. This means
that if a wave causes your boat to heave, it will most likely also cause it to pitch. This is
because the distribution of buoyancy along the hull is not constant. If the waterline is
raised some constant value, there is usually more buoyancy provided aft than there is
forward for the same change in waterline. This will cause the vessel to pitch forward to
move the center of buoyancy to the longitudinal location of the center of gravity. Sim¬
ilarly, a pitch event will cause the vessel to heave because if the stern is lifted from the
water, there is usually a loss of total buoyancy, and the boat will sink farther into the
water.
Resistance and Propulsion
Resistance is the amount of force it takes to move a body through the water. Propulsion
is the method by which you create that force.
General Resistance
In Chapter 3, we discussed drag forces on objects moving through a fluid. Specifically,
we discussed frictional and pressure drag. Ships moving on the water’s surface experi¬
ence these drag forces; however, at the air-water interface, there are other drag compo¬
nents that you have to consider. If you were to write an equation breaking up the total
resistance acting on a ship into its three main components, that equation would look
something like this:
Rtotal — Rfriction + R|
pressure
+ R\>
328 | Chapter 16: Ships and Boats
www.it-ebooks.info
We’ll describe each of these components and give you some empirical formulas in just
a moment. First, however, we want to qualify the material to follow by saying it is very
general in nature and applicable only when little detail is known about the complete
geometry of the particular ship under consideration. In the practice of ship design, these
formulas would be used only in the very early stages of the design process to approximate
resistance. That said, they are very useful for getting in the ballpark, so to speak, and
(sometimes more importantly) in performing parametric studies to see the effects of
changes in major parameters.
The first resistance component is the frictional drag on the underwater surface of the
hull as it moves through the water. This is the same as the frictional drag that we dis¬
cussed in Chapter 3. However, for ships there’s a convenient set of empirical formulas
that you can use to calculate this force:
Rfriction — (1/2) p V - S Cf
In this formula, p is the density of water, V is the speed of the ship, S is the surface area
of the underwater portion of the hull, and C f is the coefficient of frictional resistance.
You can use this empirical formula to calculate C f :
C f = 0.075 /(loglORn- 2)2
Here, Rn is the Reynolds number, as defined in Chapter 6, based on the length of the
ship’s hull. This formula was adopted in 1957 by the International Towing Tank Con¬
ference (ITTC) and is widely used in the field of naval architecture for estimating fric¬
tional resistance coefficients for ships.
To apply the formula for R friction , you’ll also have to know the surface area, S, of the
underwater portion of the hull. You can directly calculate this area using numerical
integration techniques, similar to those for calculating volume, or you can use yet an¬
other empirical formula:
S = C ws V(VL)
In this formula, V is the displaced volume, L the length of the ship, and C ws is the wetted
surface coefficient. This coefficient is a function of the ship’s shape—its breadth-to-draft
ratio—and statistically it ranges from 2.6 to 2.9 for typical displacement hull forms.
The pressure drag experienced by a ship is the same as that experienced by projectiles
as discussed in Chapter 3. Remember, this drag is due to the viscous effects causing a
region of relatively low pressure behind the ship. Quantifying this force is difficult for
ships of arbitrary geometry. We can use computational fluid dynamics algorithms to
approximate this force, but this requires detailed knowledge of the hull geometry and
a whole lot of time-consuming computations. An alternative is to rely on scale-model
Resistance and Propulsion | 329
www.it-ebooks.info
test data where results from the model test are extrapolated to approximate drag on the
full-size ship.
Just like pressure drag, wave drag is difficult to compute, and we usually rely on model
testing in practice. Wave drag is due to the energy transfer, or momentum transfer, from
the ship to the fluid, or in other words, it’s a function of the work done by the ship on
the surrounding fluid to generate the waves. The visible presence of wave resistance is
evident in the large bow wave that builds up at the front of the ship as well as the wave
system that originates at the stern of the ship as it moves through the water. These waves
affect the pressure distribution around the ship and thus affect the pressure drag, which
makes it difficult for us to separate the wave drag component from pressure drag when
performing an analysis.
When scale model tests are performed, pressure drag and wave drag are usually lumped
together in what’s known as residual resistance. Analogous to the coefficient of frictional
drag, you can determine a coefficient of residual resistance, such that:
R r — Rpressure +
= (1/2) p V 2 S C r
Here R, is the total residual resistance, and C r is the coefficient of residual resistance.
There are many resistance estimation methods available that allow you to estimate the
coefficient of residual resistance for a ship; however, they are usually presented for
specific ship types. For example, one method might give empirical formulas for C r for
destroyer-type ships, while another might give formulas for C r for large oil tankers. The
trick, of course, is to choose a method appropriate for the type of ship you are analyzing.
1 Generally, C r increases as the displacement and speed of the ship increase. A typical
range for C r for large displacement hulls is from 1.0e-3 to 3.0e-3.
While these three resistance components—friction, pressure, and wave—are the most
important for typical displacement-type ships, they aren’t the only ones. Since a ship
operates at the air-water interface, a large part of its structure is above the water surface,
exposed to the air. This means that the ship will also experience air resistance. You can
approximate this air resistance using the following formula:
R air = (1/2) p V 2 Ap C a j r
Here, C air is the coefficient of air resistance, p is the density of air, V is the speed of the
ship, and A p is the projected transverse (profile) area of the ship. C air typically ranges
from 0.6 to 1.1, depending on the type of ship. Tankers and large cargo ships tend to be
near the upper end of the range, while combatant ships tend to be near the lower end.
1. These methods are quite involved and there are far too many to discuss here, so we’ve included some references
in the Bibliography for you.
330 | Chapter 16: Ships and Boats
www.it-ebooks.info
In lieu of enough information to calculate the projected transverse area of the ship, you
can approximate it by:
A p = B 2 /2
where B is the breadth (width) of the ship.
Planing craft
Planing craft are different than displacement vessels in that when they reach their cruis¬
ing speed they are not supported by buoyancy. Unlike a super tanker, a recreational
speedboat has a much “flatter” hull form. It almost looks like a very fat foil, which it is!
When a speedboat is just sitting in the water it is fully supported by its buoyancy, just
like a tanker. However, as the boat starts moving forward, the hull is at an angle of attack
to the water. Like a super tanker, the boat is also creating a wave in front of it as it moves,
called the bow wave. However, similar to an airplane breaking the sound barrier, a
planing craft is fast enough to catch up to this wave. As it reaches its own bow wave, the
vessel will start to tilt backward. This tilting increases the resistance by virtue of in¬
creasing the angle of attack of the vessel’s hull. However, if the vessel has additional
power to overcome this increased resistance, the lift force generated from the hull foil
will begin to lift the hull out of the water. At this point, the forces in Figure 16-4 begin
to dominate.
As you can see, compared to when the boat is at rest, very little of the hull is in the water
now. This, in turn, provides positive feedback because it reduces the wetted surface area
and skin friction and allows the craft to go faster, generating more lift and reducing the
skin friction yet again. Simultaneously, the drag created by the foil moving through the
fluid increases until eventually the available power is exhausted and top speed is attained.
If we plot the resistance versus speed, it might look something like Figure 16-5.
Resistance and Propulsion | 331
www.it-ebooks.info
Notice the hump region. This is where the vessel trims aft in the transition mode and
there is a rise in the resistance. There have been embarrassing cases where a vessel,
although having enough power to make the design speed once over the humps, lacked
the power to make the transition.
Virtual mass
The concept of virtual mass is important for calculating the acceleration of a ship in a
real-time simulator. Virtual mass is equal to the mass of the ship plus the mass of the
water that is accelerated with the ship.
Back in Chapter 3 we told you about the viscous boundary layer, and we said that the
relative velocity (relative to the moving body) of the fluid particles near the moving
body’s surface is 0 at the body surface and increases to the free stream velocity as distance
from the body surface increases. Essentially, some of the fluid sticks to the body as it
moves and is accelerated with the body. Since the velocity of the fluid varies within the
boundary layer, so does the acceleration. The added mass, the mass of water that gets
accelerated, is a weighted integration of the entire mass of fluid that is affected by the
body’s acceleration.
332 | Chapter 16: Ships and Boats
www.it-ebooks.info
For a ship, the viscous boundary layer can be quite thick, up to several feet near the end
of the ship depending on its length, and the mass of water that gets accelerated is sig¬
nificant. Therefore, when doing any sort of analysis that involves the acceleration of the
ship, you need to consider added mass, too. The calculation of added mass is beyond
the scope of this book. We should also point out that, unlike mass, added mass is a tensor
—that is, it depends on the direction of acceleration. Further, added mass applies to
both linear and angular motion.
Added mass is typically expressed in terms of an added mass coefficient, which equals
the added mass divided by the mass of the ship. Some methods actually integrate over
the actual hull surface, while others approximate the hull as an ellipsoid with proportions
matching the ship’s. Using this approximation, the ellipsoids length corresponds to the
ship’s length while its width corresponds to the ship’s breadth. For longitudinal motion
(that is, linear motion along an axis parallel to the ship’s length), the added mass coef¬
ficient varies nearly linearly from 0.0 at a breadth-to-length ratio of 0 (the ship is in¬
finitely thin) up to 0.5 at a breadth-to-length ratio of 1 (a sphere).
When the added mass coefficient is expressed as a percentage of the ship’s mass, virtual
mass can be calculated as mv = m (1 + xa), where m is mass, and xa is the added mass
coefficient—for example, 0.2 for 20%. For typical displacement ship proportions, the
longitudinal added mass ranges from about 4% to 15% of the mass of the ship. Con¬
servative estimates generally use 20%.
Guidance speeds
To provide some guidance, Table 16-1 provides common ship types and appropriate
speed ranges. This will help guide you in properly simulating the resistance of your
vessel.
Table 16-1. Some vessels and their speeds
1 Vessel type
Speed (knots)
Horsepower (hp) 1
Aircraft carrier
31.5
260,000
Cruiser
30
80,000
Oil tanker
15-20
20,000-60,000
Containership
21
100,000
200-foot yacht
15.5
4,000
35-foot recreational boat
30
420
35-foot speedboat
70
1,200
40-foot sailboat
8.5
N/A
Note that at a certain speed, for non-planing hulls, there is a theoretical limit to how
fast a boat can go. This speed is called the hull speed. At the hull speed the bow and stern
waves reinforce each other, and there is a rise in wave-making resistance. This can be a
Resistance and Propulsion | 333
www.it-ebooks.info
barrier for some fuller hulls. Note that the speed for the 40-foot sailboat is the hull speed
of a 40-foot full-formed (not slender) hull. We can calculate the hull speed with the
following formula:
Vhull ~ 1.34 xjLWL
Where V hull is the hull speed in knots and L wl is the length on the water line. Some
modern displacement boats, especially racing kayaks, can exceed their hull speed if they
have very fine ends, long hulls with narrow beams, or some other optimized hull form.
Propulsion
There are a variety of methods to propel a boat through the water. The oldest is by way
of sails using the wind. The physics of sailing could fill several chapters on its own, so
we won’t go into detail here. If you are interested in that topic, The Physics of Sailing
Explained (Sheridan House), by Bryon Anderson, is a good starting point. You can also
check out the airplane chapter in this book and note that a sail is just a vertical wing.
We will say this: if you choose to put a sailing vessel in your simulation, make sure to
remember the golden rule! You cannot generally sail within 45 degrees of the direction
the wind is coming from.
There are many different kinds of propulsion: rudders, shafted fixed-pitch propellers,
shafted CPP propellers, azimuthing thrusters, propellers in nozzles, water jets, pump
jets, contra-rotating, and the very odd Voith Schneider type. We can’t get into all of them
here, but the aforementioned list should get you started if you are interesting in mod¬
eling the specifics of different propulsor types.
The maj or propulsion relationship for your simulation would be the thrust-to-propeller
RPM or thrust-to-throttle curve. The fly in the ointment here is that thrust is also de¬
pendent on forward speed. As the boat moves faster and faster through the water, the
inflow velocity of the water into the propeller is higher and higher. This means that the
difference between the velocity of the intake and the velocity of the output is smaller,
reducing the amount of thrust. In general, the thrust curve versus boat speed will look
like Figure 16-6.
334 | Chapter 16: Ships and Boats
www.it-ebooks.info
An important physical phenomenon concerning propulsion that you may want to in¬
corporate is cavitation. Cavitation occurs when a propeller is moving fast enough that
the low-pressure side of the blade starts spontaneously creating vapor bubbles. These
bubbles exist for a short while, and then as the propeller turns, the static pressure
changes. This higher static pressure causes the bubbles to collapse violently. This collapse
is so fast and furious that it can cause metal erosion at a high rate. It will eat away at a
propeller until it is no longer producing thrust. It is also very noisy. That is why sub¬
marine propellers are shaped very differently than other ships’ propellers. They seek to
limit cavitations so they’re not heard by enemy vessels. The damage caused by cavitation
also creates a speed limit on RPMs for a propeller. Cavitation is a real-life phenomenon
you can exploit to penalize the player for driving around at high speed all the time.
Maneuverability
Another aspect of ships and boats that is often oversimplified is maneuverability. Ma¬
neuverability is also a very complex topic whose numerical simulation is beyond the
current realm of real-time simulation. However, with some simplifications and as¬
sumptions, it can be more accurately modeled than if you do not know the underlying
framework. Almost all vessels maneuver by way of two methods: rudders or thrust
vectoring. The users generally won’t care about the differences, so you can model both
by angling the thrust vector off-center.
Maneuverability | 335
www.it-ebooks.info
Rudders and Thrust Vectoring
Although rudders and thrust vectoring have the same result, there are some important
differences. A thrust vectoring system, like a jet boat, can steer only when the vessel is
producing thrust. A rudder, on the other hand, works only when the vessel has forward
speed. If the boat isn’t moving forward with enough speed, then the rudder can’t produce
a turning moment.
If you keep in mind those differences, you can model both systems the same way. The
most important thing to keep in mind when modeling larger ships in your games is that
they take significant time to respond to control inputs. Figure 16-7 tracks the heading
of a ship over time during what is called the 10/10 maneuver.
Figure 16-7. 10/10 zig-zag test
A vessel is moving in a straight line, and the rudder is put over 10 degrees in one di¬
rection. Once the vessel’s heading changes 10 degrees, the rudder is moved to the op¬
posite side at the same angle. The initial turning time is the time it took for the vessel
to change its heading 10 degrees. As large ships have enormous momentum, they will
continue to turn even though the rudder is in the opposite direction. The maximum
deviation from the original heading minus the 10 degrees at which the rudder was
flipped is called the overshoot angle. The size of this angle is one measure of how slow
the vessel is to respond to the helm. For larger ships, this can be between 15 degrees
when light and 45 degrees when loaded with cargo.
The time to check yaw is the time in seconds it takes for the overshoot angle to be
achieved and the vessel to start changing its heading again. This is repeated for the other
side to detect any bias a vessel may have for turning in a particular direction. The moral
of the story is that for anything other than a high-speed small craft, boats and ships can
take a significant amount of time to respond to the helm. Your simulation should strive
to reproduce a turning ability that matches Figure 16-7 for extra realism.
336 | Chapter 16: Ships and Boats
www.it-ebooks.info
A special kind of thrust vectoring is called throttle steering. Imagine that a boat has two
engines. If one is run in forward and the other in reverse, the vessel will turn quite
rapidly. For a twin-engine vessel operating in close quarters, the rudders are often cen¬
tered and the vessel maneuvered solely by altering the throttle settings of the two props.
Propeller walk
Another interesting maneuvering phenomenon that is closely related to thrust vectoring
is called propeller walk, or prop walk. This is especially important for vessels with only
one propeller moving in tight spaces. The cause of propeller walk is related to the fact
that most propellers are installed at an angle to the horizon. This angle causes the thrust
to be greater when the blades are moving down than when the blades are moving up¬
ward. In a propeller that turns clockwise, this creates a push to the right.
In forward gear the rudder is very effective at countering the propeller walk, but in
reverse the rudder is much less effective, making the propeller walk much more no¬
ticeable. This can add a significant amount of realism when you are simulating vessels
in docking maneuvers.
Maneuverability | 337
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 17
Cars and Hovercraft
What cars and hovercraft have in common is that they operate in an essentially 2D
manner. Unless they have jumped a ramp, both vehicles remain on the ground or water
plane. In this chapter we’ll discuss the forces behind each vehicle’s method of travel and
discuss how to accurately model them in your simulations.
Cars
In the following sections we want to discuss certain aspects of the physics behind au¬
tomobile motion. Like the previous four chapters, the purpose of this chapter is to
explain, by example, certain physical phenomena. We also want to give you a basic
understanding of the mechanics involved in automobile motion in case you want to
simulate one in your games. In keeping with the theme of this book, we’ll be talking
about mechanics in the sense of rigid-body motion, and not in the sense of how an
internal combustion engine works, or how power is transferred through the transmis¬
sion system to the wheels, etc. Those are all internal to the car as a rigid body, and we’ll
focus on the external forces. We will, however, discuss how the torque applied to the
drive wheel is translated to a force that pushes the car along.
Resistance
Before we talk about why cars move forward, let’s talk about what slows them down.
When a car drives down a road, it experiences two main components of resistance that
try to slow it down. The first component is aerodynamic drag, and the second is called
rolling resistance. The total resistance felt by the car is the sum of these two components:
Rtotal = Rair + ^rolling
339
www.it-ebooks.info
The aerodynamic drag is primarily skin friction and pressure drag similar to that ex¬
perienced by the projectiles discussed in Chapter 6, and the planes and boats discussed
in earlier chapters. Here again, you can use the familiar drag formula:
R air = (1/2) p V 2 S p C d
Here p (rho) is the mass density of air, V is the speed of the car, S p is the projected frontal
area of the car normal to the direction of V, and C d is the drag coefficient. Typical ranges
of drag coefficients for different types of vehicles are 0.29 to 0.4 for sports cars, 0.43 to
0.5 for pickup trucks, 0.6 to 0.9 for tractor-trailers, and 0.4 to 0.5 for the average economy
car. Drag coefficient is a function of the shape of the vehicle—that is, the degree of
boxiness or streamline. Streamlined body styles have lower drag coefficients; for ex¬
ample, the Chevy Corvette has a low drag coefficient of 0.29, while the typical tractor-
trailer without fairings has a high drag coefficient of up to 0.9. You can use these coef¬
ficients in your simulations to tune the behavior of different types and shapes of vehicles.
When a tire rolls on a road, it experiences rolling resistance, which tends to retard its
motion. Rolling resistance is not frictional resistance, but instead has to do with the
deformation of the tire while rolling. It’s a difficult quantity to calculate theoretically
since it’s a function of a number of complicated factors—such as tire and road defor¬
mation, the pressure over the contact area of the tire, the elastic properties of the tire
and road materials, the roughness of the tire and road surfaces, and tire pressure, to
name a few—so instead you’ll have to rely on an empirical formula. The formula to use
is as follows:
^rolling — C r W
This gives you the rolling resistance per tire, where w is the weight supported by the
tire, and C r is the coefficient of rolling resistance. C r is simply the ratio of the rolling
resistance force to the weight supported by the tire. Luckily for you, tire manufacturers
generally provide the coefficient of rolling resistance for their tires under design con¬
ditions. Typical car tires have a C r of about 0.015, while truck tires fall within the range
of 0.006 to 0.01. If you assume that a car has four identical tires, then you can estimate
the total rolling resistance for the car by substituting the total car weight for w in the
preceding equation.
Power
Now that you know how to calculate the total resistance on your car, you can easily
calculate the power required to overcome the resistance at a given speed. Power is a
measure of the amount of work done by a force, or torque, over time. Mechanical work
done by a force is equal to the force times the distance an object moves under the action
of that force. It’s expressed in units such as foot-pounds. Since power is a measure of
340 | Chapter 17: Cars and Hovercraft
www.it-ebooks.info
work done over time, its units are, for example, foot-pounds per second. Usually power
in the context of car engine output is expressed in units of horsepower, where 1 horse¬
power equals 550 ft-lbs/s.
To calculate the horsepower required to overcome total resistance at a given speed, you
simply use this formula:
P = (RtotalV)/550
Here P is power in units of horsepower, and i? total is the total resistance corresponding
to the car’s speed, V. Note, in this equation Jf total must be in pound units and V must be
in units of feet per second.
Now this is not the engine output power required to reach the speed V for your car;
rather, it is the required power delivered by the drive wheel to reach the speed V. The
installed engine power will be higher for several reasons. First, there will be mechanical
losses associated with delivering the power from the engine through the transmission
and drive train to the tire. The power will actually reach the tire in the form of torque,
which given the radius of the tire will create a force f' w that will overcome the total
resistance. This force is calculated as follows:
Fw = T w /r
Here F w is the force delivered by the tire to the road to push the car along, T w is the
torque on the tire, and r is the radius of the tire. The second reason the installed engine
power will be greater is because some engine power will be transferred to other systems
in the car. For example, power is required to charge the battery and to run the air
conditioner.
Stopping Distance
Under normal conditions, stopping distance is a function of the braking system and
how hard the driver applies the brakes: the harder the brakes are applied, the shorter
the stopping distance. That’s not the case when the tires start to skid. Under skidding
conditions, stopping distance is a function of the frictional force that develops between
the tires and the road, in addition to the inclination of the roadway. If the car is traveling
uphill, the skidding distance will be shorter because gravity helps slow the car, while it
will tend to accelerate the car and increase the skidding distance when the car is traveling
downhill.
There’s a simple formula that considers these factors that you can use to calculate skid¬
ding distance:
d s = V 2 / [2g ( p cos cp + sin cp)]
Cars | 341
www.it-ebooks.info
Here d s is the skidding distance, g the acceleration due to gravity, p the coefficient of
friction between the tires and road, V the initial speed of the car, and cp the inclination
of the roadway (where a positive angle means uphill and a negative angle means down¬
hill). Note that this equation does not take into account any aerodynamic drag that will
help slow the car down.
The coefficient of friction will vary depending on the condition of the tires and surface
of the road, but for rubber on pavement the dynamic friction coefficient is typically
around 0.4, while the static coefficient is around 0.55.
When calculating the actual frictional force between the tire and road, say in a real-time
simulation, you’ll use the same formula that we showed you in Chapter 3:
F f = p W
Here F f is the friction force applied to each tire, assuming it’s not rolling, and W is the
weight supported by each tire. If you assume that all tires are identical, then you can use
the total weight of the car in the preceding formula to determine the total friction force
applied to all tires.
Steering
When you turn the steering wheel of a car, the front wheels exert a side force such that
the car starts to turn. In terms of Euler angles, this would be yaw, although Euler angles
aren’t usually used in discussions about turning cars. Even if the car’s speed is constant,
it experiences acceleration due to the fact that its velocity vector has changed direction.
Remember, acceleration is the time rate of change in velocity, which has both magnitude
and direction.
For a car to maintain its curved path, there must be a centripetal force (“center seeking”
in Greek) that acts on the car. When riding in a turning car, you feel an apparent
centrifugal acceleration, or force directed away from the center of the turn. This accel¬
eration is really a result of inertia, the tendency of your body and the car to continue on
their original path, and is not a real force acting on the car or your body. The real force
is the centripetal force, and without it your car would continue on its straight path and
not along the curve.
One of the most important aspects of racing is taking turns as fast as possible but without
losing control. The longer you can wait to decelerate for the turn, and the sooner you
can start accelerating again, the higher your average speed will be. It is interesting to
note that when people ride in racecars, what really surprises them isn’t the acceleration
but the massive decelerations they can create through braking forces. A Formula One
car regularly experiences decelerations of 4g, with 5-6g being the extreme value for
certain race courses. Most road-legal sports cars can achieve about lg of braking force.
This allows the racecars to maintain speed until just entering the corner.
342 | Chapterl7:CarsandHovercraft
www.it-ebooks.info
When you turn the steering wheel in a car, the tires produce a centripetal force toward
the center of the curve via friction with the surface of the road. It follows that the max¬
imum static frictional force between the tires and the road must exceed the required
centripetal force. Mathematically, this takes on the following inequality:
p S N > mv 2 /r
Centripetal acceleration is the square of the tangential velocity, v, divided by the radius
of the turn, r. This multiplied by mass of the vehicle, m, gives the force required to make
a turn. The available force is the static coefficient of friction times the normal force, N.
Rewriting this formula, we can develop a simplified maximum cornering speed as fol¬
lows:
If this speed is exceeded, the cornering force will exceed the static coefficient of friction
and the tires will begin to slide. It may be tempting to replace the normal force, N, with
the weight of the vehicle (mass times gravity) and cancel out the mass of the car, but the
normal force may not always be simplified as such, as we’ll discuss in a moment. Also,
in real life, the weight of a vehicle is rarely evenly distributed among all four tires and
is definitely not when the car is acceleration or decelerating. In general, acceleration
causes a weight shift to the aft tires, and deceleration causes weight to shift to the forward
tires. This is important because, depending on where the weight is, the front or back
tires will tend to have less available friction. If one or the other sets of tires begins to
slide, the car will either understeer or oversteer (see Figure 17-1).
Figure 17-1. Understeer on right, oversteer on left
Cars | 343
www.it-ebooks.info
If the front wheels slide, the car understeers and the arc is larger than the driver intended.
This is commonly caused by traveling too fast through a corner and trying to take the
corner too tightly. However, if the driver breaks hard or even just lets off the gas, there
will be a weight shift forward. This will keep the forward wheels from slipping but if
too aggressive will cause the rear wheels to slip as weight is transferred away from them.
This causes the car to turn more than the driver intended, and can even result in a spin.
These two conditions limit the speed at which a car can complete a turn and also the
amount of deceleration the car can handle once the turn is initiated.
To increase the limit speed, v limit , we must increase the normal force. We could do so by
increasing the car’s mass, but this would have negative effects on the car’s ability to
accelerate or decelerate. A better solution is the use of aerodynamic features to create
what is called downforce. You may have seen racecars or even street cars with large wing¬
like features called spoilers on their trunks. Formula 1 cars also have wing-like appen¬
dages on the front of the car. These are like the wings of an airplane but inverted so that
instead of pulling the vehicle up, they actually push the vehicle down. These wings,
therefore, increase the normal force, and their effects are proportional to speed so that
more downforce is available at higher speeds...just when you need it. In fact, some very
fast cars create so much downforce that if the road were inverted, the car would remain
glued to the pavement and be able to drive upside-down.
The trade-off for this increased cornering limit speed is increased drag caused by the
airfoils. This limits the top speed of the vehicle in areas of road that have no corners. It
is good practice to allow users to select an angle of attack for their vehicle’s airfoils. This
forces them to choose between the option of higher top speed with slower cornering or
faster cornering with lower top speed.
You may notice that on some raceways, the corners are not flat. This is called roadway
bank or superelevation. In a corner where the car would normally skid, the supereleva¬
tion helps keep the car in the turn because as the car is inclined, a force component
develops that acts toward the center of curvature of the turn (see Figure 17-2).
Figure 17-2. Superelevation
344 | Chapter 17: Cars and Hovercraft
www.it-ebooks.info
The following simple formula relates the superelevation angle of a roadway to the speed
of the car and the coefficient of friction between the tires and road:
tan (p = V t 2 /(g r) - p s
Here, cp is the superelevation angle (as shown in Figure 17-2), V t is the tangential com¬
ponent of velocity of the car going around the turn, g is the acceleration due to gravity,
r is the radius of the curve, and p s is the static coefficient of friction between the tires
and the road. If you know cp, r, and p, then you can calculate the speed at which the car
will begin to slip out of the turn and off the road.
Vehicle dynamics is a complex field, and if you are interested in a highly realistic driving
simulation game, we recommend reading up on the subject. A good starting place is
Fundamentals of Vehicle Dynamics by Thomas Gillespie (Society of Automobile Engi¬
neers).
Hovercraft
Hovercraft, or air cushion vehicles (ACVs), have made their way into a video game or
two recently. Their appeal seems to stem from their futuristic aura, high speed, and
levitating ability, which lets them go anywhere. In real life, hovercraft have been around
since the 1950s and have been used in combat, search and rescue, cargo transport,
ferrying, and recreational roles. They come in all shapes and sizes, but they all pretty
much work the same, with the basic idea of getting the craft off the land or water to
reduce its drag. In Chapter 9, we touched on what forces you have to model when
considering hovercraft, and now we’ll talk about them in more detail.
How Hovercraft Work
The first hovercraft designs pumped air through an annular nozzle around the periphery
of the craft (see Figure 17-3). Large fans are used to feed the air through the nozzle under
the craft. This jet of air creates a region of relatively high pressure over the area under¬
neath the craft, which results in a net lifting force. The lifting force must equal the weight
of the craft if it is to attain hovering flight. This sort of lifting is known as aerostatic lift.
The hover height is limited by the amount of power available and the lifting fan’s ability
to pump enough air through the nozzle: the higher the hover height, the greater the
power demand.
Hovercraft | 345
www.it-ebooks.info
» T » T T
Fan
Nozzle
Jet •
Y\\ \ * } i * /
J y -..Air cushion y y
Ground
t t ♦ t »
Fan
* /
V * <
/4/r cushion
Plenum chamber
\\ \ k ;
S/r/rf
Ground
Figure 17-3. Hovercraft configurations
This approach proved impractical because hover heights were very limited and made
the clearance between the hard structure of the craft and the ground (or water) too small
to overcome all but the smallest obstacles. The solution to this problem was to fit a
flexible skirt around the craft to contain the air cushion in what’s called the plenum
chamber (see Figure 17-3). This approach extended the clearance between the ground
and the hard structure of the craft significantly even though the gap between the bottom
of the skirt and the ground was very small. This is the basic configuration of most
hovercraft in operation today, although there are all sorts of skirt designs. Some of these
skirts are simple curtains, while others are sophisticated pressurized bag and finger
arrangements. The end result is that hovercraft fitted with skirts can clear relatively large
obstacles without damage to their hard structure, and the skirt simply distorts and
conforms to the terrain over which the craft operates.
The actual calculation of the aerostatic lift force is fairly complicated because the pres¬
sure distribution within the air cushion is nonuniform and because you must also take
into account the performance of the lift fan system. There are theories available to treat
both the annular jet and plenum chamber configurations, but they are beyond the scope
of this book. Besides, for a game simulation, what’s important is that you realize that the
lift force must equal the weight of the craft in order for it to maintain equilibrium in
hovering flight.
346 | Chapter 17: Cars and Hovercraft
www.it-ebooks.info
Ideally, the ability of a hovercraft to eliminate contact with the ground (or water) over
which it operates means that it can travel relatively fast since it no longer experiences
contact drag forces. Notice we said ideally. In reality, hovercraft often pitch and roll,
causing parts of the skirt to drag, and any obstacle that comes into contact with the skirt
will cause more drag. At any rate, while eliminating ground contact is good for speed,
it’s not so good for maneuverability.
Hovercraft are notoriously difficult to control since they glide across the ground. They
tend to continue on their original trajectory even after you try to turn them. Currently,
there are several means employed in various configurations for directional control.
Some hovercraft use vertical tail rudders much like an airplane, while others actually
vector their propulsion thrust. Still others use bow thrusters, which offer very good
control. All of these means are fairly easy to model in a simulation; they are all simply
forces acting on the craft at some distance from its center of gravity so as to create a
yawing moment. The 2D simulation that we walked you through in Chapter 9 shows
how to handle bow thrusters. You can handle vertical tail rudders as we showed you in
Chapter 15.
Resistance
Let’s take a look now at some of the drag forces acting on a hovercraft during flight. To
do this, we’ll handle operation over land separately from operation over water since
there are some specific differences in the drag forces experienced by the hovercraft.
When a hovercraft is operating over smooth land, the total drag acting against the
hovercraft is aerodynamic in nature. This assumes that drag induced by dragging the
skirt or hitting obstacles is ignored. The three components of aerodynamic drag are:
• Skin friction and viscous pressure drag on the body of the craft
• Induced drag when the craft is pitched
• Momentum drag
In equation form, the total drag is as follows:
Rtotal = ^viscous T ^induced + ^momentum
The first of these components, the viscous drag on the body of the craft, is the same sort
of drag experienced by projectiles flying through the air, as explained in Chapter 6. This
drag is estimated using the by-now-familiar formula:
^viscous = (1/2) p V - Sp Cjj
Hovercraft | 347
www.it-ebooks.info
Here p is the mass density of air, V the speed of the hovercraft, S p the projected frontal
area of the craft normal to the direction of V, and C d the drag coefficient. Typical values
of C d for craft in operation today range from 0.25 to 0.4.
The next drag component, the induced drag, is a result of the craft assuming a pitched
attitude when moving. When the bow of the craft pitches up by an angle t, there will
be a component of the aerostatic lift vector that acts in a direction opposing V. This
component is approximately equal to the weight of the craft times the tangent of the
pitch angle:
^induced ~ W (tan t)
Finally, momentum drag results from the destruction of horizontal momentum of air,
relative to the craft entering the lift fan intake. This component is difficult to compute
unless you know the properties of the entire lifting system such that the mass flow rate
of air into the fan is known. Given the mass flow rate, k mnmcntLlm is equal to the mass flow
rate times the velocity of the craft:
^momentum = (dm fan /dt) V
Mass flow rate is expressed in units such as kg/s, which when multiplied by velocity in
m/s yields N.
In Chapter 9 we mentioned that it is beneficial to have the center of this drag be behind
the center of gravity. This gives directional stability, as the vessel will tend to try to point
into the apparent wind. Consider if the vessel yaws some angle, the forward velocity is
creating a wind that is now hitting the side of the hovercraft. If the center of effort of
this force is forward of the center of gravity, it will want to yaw the vessel more, increasing
the side force and causing even greater drag until the vessel spins 180 degrees! If the
center of effort of the wind is aft of the center of gravity, the vessel will spin back into
the wind. You don’t want the center of effort so far aft that it is too difficult to turn the
vessel, but it is generally better to have the vessel naturally straighten out than to require
constant steering input to maintain a steady course. This is also true of sailboats and is
called weather helm. This also occurs in airplanes. The solution in airplanes is similar
to that of hovercraft: to fit tail fins that move the center of the area aft, increasing its
directional stability. If you choose to model damage in your simulation, the loss of a tail
fin will cause the vessel to be very difficult to control.
In addition to these three drag components, hovercraft will experience other forms of
resistance when operating over water. These additional components are wave drag and
wetted drag. The equation for total drag can thus be revised for operation over water as
follows:
R. i ■
348 | Chapter 17: Cars and Hovercraft
www.it-ebooks.info
When a hovercraft operates over water, its air cushion creates a depression in the water
surface due to cushion pressure (see Figure 17-4). At zero to low speeds, the weight of
this displaced volume of water is equal to the weight of the craft, just as if the craft were
floating in the water supported by buoyancy. As the craft starts to move forward, it tends
to pitch up by the bow. When that happens, the surface of the water in the depressed
region is approximately parallel to the bottom of the craft. As speed increases, the de¬
pression is reduced and the pitch angle tends to decrease.
Figure 17-4. Hovercraft over water
Wave drag is a result of this depression and is equal to the horizontal components of
pressure forces acting on the water surface in the depressed region. As it turns out, for
small pitch angles and at low speeds, wave drag is on the same order of magnitude as
the induced drag:
Rwave ~ W (tan t)
Since wave drag is proportional to the size of the depression, it tends to be highest at
low speeds and decreases at higher operational speeds. If you were to plot the wave drag
curve as a function of speed for a typical hovercraft, you’d find that it is not a straight
or even parabolic curve, but rather it has a hump in the curve at the lower speed range,
as illustrated in Figure 17-5.
Hovercraft | 349
www.it-ebooks.info
There are several theoretical treatments of wave drag in the literature that aim to predict
the speed at which this hump occurs along with its magnitude. These theories indicate
that the hump depends on the planform geometry of the hovercraft, and it tends to
occur at speeds in the range of *JgL 1 2 to *jgL , where g is the acceleration due to gravity
and L is the length of the air cushion. In practice, the characteristics of a particular
hovercraft’s wave drag are usually best determined through scale model testing.
The so-called wetted drag is a function of several things:
• The fact that parts of the hull and skirt tend to hit the water surface during flight
• The impact of spray on the hull and skirt
• The increase in weight as the hovercraft gets wet and sometimes takes on water
Wetted drag is difficult to predict, and in practice model tests are relied on to determine
its magnitude for a particular design. It’s important to note, however, that this tends to
be a significant drag component, sometimes accounting for as much as 30% of the total
drag force.
Steering
In Chapter 9, the hovercraft was steered using a bow thruster that pushed transversely
forward of the center of gravity. In reality, most hovercraft are steered by vectoring the
thrust of the propulsion fan via rudders attached directly aft of the fans. This can be
modeled by angling the propulsive thrust.
The most important characteristic to remember about steering hovercraft is that they
will not turn like a car or boat. The hovercraft, because it has lower friction with its
environment, will take longer to turn and tends to continue in the direction it was
heading while rotating. Once rotated, the thrust acts along a new vector. One possible
maneuver in a hovercraft is to quickly rotate the vessel and then shut down the pro-
350 | Chapter 17: Cars and Hovercraft
www.it-ebooks.info
pulsive thrust. This would allow you to travel in one direction and point in another for
as long as your momentum carries you. This could be very useful in strafing enemies
or just racking up style points.
Hovercraft |
351
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 18
Guns and Explosions
One of the most widely successful video game genres is the venerable first-person
shooter. Ever since the breakthrough games of Wolfenstein 3D and Doom, the first-
person shooter has received the lion’s share of research and development budgets. It is
amazing that the physics of aiming a gun and of a bullet traveling through the air are
rarely modeled accurately. In general, game designers treat guns like laser beams so that
wherever you point them, the bullet goes in an infinitely straight line. In this chapter
we’ll discuss how to more accurately model both aiming and the trajectory of bullets,
which is known as ballistics.
Projectile Motion
There are actually four subtopics of ballistics. Internal ballistics is the study of what
happens to the bullet inside the barrel of the gun; transitional ballistics is the study of
what happens as the bullet exits the barrel. Once the bullet has fully exited the barrel, it
is in the realm of external ballistics. At this point the only acceleration is that of gravity,
and the same forces discussed in Chapter 6 take over. The last topic is terminal ballis¬
tics, which is the study of what happens when the bullet hits its target. The last two topics
are the ones we’ll discuss here. The other phases are more important to firearm man¬
ufacturers and not so much to the shooter. If you don’t recall the material in Chap¬
ter 6, we highly recommend that you review it before continuing.
While we aren’t very concerned about what happens in the barrel of the gun, there are
a few tidbits we do need to know about the system. The first is where the barrel is pointed.
This is referred to as the gun’s aim, and is almost universally controlled by where the
mouse is on the screen.
353
www.it-ebooks.info
The next is the initial velocity of the bullet. The bullet here refers to the actual metal
projectile that leaves the barrel; the thing that you load into the gun is called a round
and contains a casing, gunpowder, a primer, and, of course, the bullet. The initial velocity
is usually measured just after the bullet leaves the muzzle (the end of the barrel) and is
appropriately called the muzzle velocity. Every kind of ammunition is tested at the fac¬
tory and given a muzzle velocity. You can add realism to your game by giving different
ammunition different muzzle velocities. This way, a handgun round won’t have the same
range as a rifle round. Ammunition also comes with a bullet weight measured in either
grams or grains. One grain is equal to 0.0648 grams and is an old unit based on the
weight of a single seed of wheat!
Last, but not least, we need some approximation for the way air resistance will affect
the flight of a bullet. This is where things start to get interesting for people studying
ballistics, but we’ll stay away from exotic aerodynamics and use our existing drag model.
First, we should review the current state of first-person-shooter physics.
Firearms in games present a unique problem to the game developer. If you have ever
been to the firing range, you know that in reality it takes a good deal of practice and
concentration to hit a target reliably. Considering that target shooting is hard enough
to be an Olympic sport under very controlled circumstances, the ability for in-game
characters to spring from cover and shoot five enemies with five bullets is somewhat
superhuman. We have all played games where you find yourself shooting a target very
far away, and the procedure is as simple as pointing the crosshair where you want the
bullet to go and clicking the mouse button. In reality, the skill needed to get a bullet
weighing a few grams to hit something a few hundred meters away is so complicated it
is amazing that anyone does it with regularity. For those developers wishing to actually
model firearm performance in their game, there is a double-edged sword to consider.
The physics of what happens to the bullet in its flight are not simple to boil down.
However, the behavior of the bullet as it flies downrange is important in the practical
art of marksmanship. There has been considerable work done to find a way to compare
ammunition so that a hunter or marksman can predict the performance of a particular
ammunition. The result is a pseudophysical factor called the ballistic coefficient (BC).
The BC is a ratio that determines the ability of a particular bullet to retain its downrange
velocity compared to some standard bullet. The most common form is that of the G1
reference projectile. However, this number has limited use, as it does not take into
account modern bullet shapes that provide very low drag. There are updated models
whose designations are G2, G3, and ECT. If you are interested in the details of highly
accurate ballistic modeling, there are a few free programs that can provide you in-depth
models such as Remington’s Shoot! and the GNU Ballistics program. Given that most
first-person shooters don’t yet include wind effects or bullet drop, we’ll limit ourselves
to a simplified method of turning the existing parameters in the Chapter 6 projectile
algorithm.
354 | Chapter 18: Guns and Explosions
www.it-ebooks.info
Taking Aim
When discussing aim, we’ll primarily be talking about rifles or carbines here, as hand¬
guns are not usually used for long-distance shooting. Similarly, shotguns, given that
they fire many small projectiles, are generally not carefully aimed but instead pointed.
Both of these weapons are what are known as point-blank weapons. Rifles have a point-
blank range, which we will discuss, but for handguns and shotguns this really refers to
the fact that at the ranges where these weapons are effective, you can reasonably expect
to hit where you the point the gun. That is not to say that these weapons don’t need to
be aimed to be effective, but in the fast-paced combat central to most first-person
shooters it would be tedious to have the player use the sights on a handgun to effectively
hit anything. Instead, most games use a “shoot from the hip” or free aim model, where
the gun is not even in line with the camera. The careful programmer might still check
to see if the target is within the effective range of the bullet before counting it as a hit
even when auto-aiming in this manner. The effective range is the distance the bullet can
travel before hitting the ground. Determining this is a straightforward application of
the equations in Chapter 2 and Chapter 6.
To discuss firearms, we’ve adapted the code from the Cannon2 example in Chapter 6
to run in a Java program called Marksman. In our example, the player is looking through
a scope at a target of 1 meter by 1 meter. We’ve provided him an adjustable level of zoom
so that as the range increases, he can still see the target. The aiming point is shown as
an empty circle, and the bullet hole as a black filled circle. The method we’ve used to
determine where the user is aiming in the model world converts the pixel-based location
of the mouse to a coordinate in the model world. This distance and the range are then
used to find the angles required to aim the gun. The code to do this is shown next, where
alp and gm are the angles of inclination and bearing. These are measured from hori¬
zontal and the line of view:
alp = 90-Math. toDegrees(Math.atan(((200-airiY)*(targetH/(drawH)))/range));
gmm = Math.toDegrees(Math.atan(((200-almX)*(targetH/(drawH)))/range));
where targetH/drawH is just the ratio of the target height to the height of the target in
pixels on the screen. This allows the mouse coordinate given in pixels to be converted
to meters. The arctangent then converts the ratio of these distances to an angle for the
gun. The constant 200 refers to the pixel coordinate system, which is 200 pixels away
from the center of the target. If you were to adapt this for a full 3D rendering system,
you could remove a lot of these conversions we have to do.
Taking Aim | 355
www.it-ebooks.info
At the 10-meter range shown in Figure 18-1, we are within point-blank range and the
bullet hole lines up with the sights.
Figure 18-1. Point-blank range
In Figure 18-2, with the target at 100 meters, we see that for some reason the bullet is
not hitting the target where we have pointed the weapon. At 300 meters, the bullet hole
isn’t even on the target. You can see that by simply accounting for projective motion
and idealized drag, we are already having trouble hitting the bull’s eye. The process by
which these differences are accounted for is called zeroing the sights.
356 | Chapter 18: Guns and Explosions
www.it-ebooks.info
Figure 18-2. Hitting lower than expected
Zeroing the Sights
The idea of zeroing the sights is probably the most important thing to model if you want
to have realistic gunplay in your game. As we mentioned before, when players are run¬
ning from room to room, they probably don’t want to be thinking about the wind and
the range. However, for a hunting simulation or a sniper game, it may be appropriate
to introduce this.
When a person is looking through a scope, her body and the rifle become a rigid body
so that to change the aim of the weapon, she must rotate her entire body. This is con¬
venient for us, because the player generally controls the shooter’s position with his left
hand via the keyboard and the direction of aim with the right hand. Other methods of
aiming, briefly described previously as free aim, don’t really have a counterpart in the
real world, so we’ll be limiting ourselves to this solid-body aiming.
Bullet drop: Gravity and air resistance
If you are aiming a rifle horizontally, you might expect that the bullet leaves the muzzle
horizontally, and that gravity and air resistance cause it to drop from there. Figure 18-3
Taking Aim | 357
www.it-ebooks.info
shows a rifle and scope combination that is mounted perfectly parallel. Ignoring all other
factors for a minute, we see that the bullet will never hit where the scope is pointed. It
will always be a few centimeters low.
Figure 18-3. Zero-elevation scope
By adjusting the elevation control of the scope, we can make the rifle hit where the scope
is pointed. (See Figure 18-4.) The range at which the bullet will cross the line defined
by the scope is called the zero range. If a target is at the zero range, you simply point the
crosshairs and pull the trigger.
Scope Axis
—---—
Barrel Axis
Zero Ranae
Figure 18-4. Scope with elevation
If we remove the target and graph the trajectory, it would look something like
Figure 18-5.
Here we can see there is a second point in which the bullet will cross the line made by
the scope’s crosshairs. This is called the far zero, or second zero. This is generally where
the thing you will be shooting at will be. This diagram is often available from ammu-
358 | Chapter 18: Guns and Explosions
www.it-ebooks.info
nition suppliers. More often, they provide ballistic tables of downrange heights. It is
important when using these charts that you remember they assume the scope to be
horizontal and that the bullet starts with a negative height below the scope line.
Table 18-1 shows data from Remington’s website and assumes the rifle was zeroed at
100 yards; it tells you how far the bullet is below the horizontal at every 50 yards past
that 0. You can tune the drag coefficient in the projectile simulation to match these
elevations.
Table 18-1. Long-range trajectory Remington Express .45-70 Govt
Range in yards 100 150 200 250 300 400 500
Drop in inches 0 -4.6 -13.8 -28.6 -50.1 -115.47 -219.1
From this table you can see that the bullet drop is over two feet at 250 yards. This means
that if the target is at 250 yards and your scope is zeroed at 100 yards, then you would
have to aim two feet above your target to hit it. That might be high enough that you
can’t even see the target in the scope anymore!
To counteract this problem, most rifles come with scopes that have elevation adjust¬
ments. This is a knob on the side of the scope that can be rotated to discrete settings
called clicks. Most scopes use a 1/4 minute of angle adjustment per click although some
use 1/8,1/2, or even full minutes. A minute of angle is simply 1/60 of a degree. Therefore,
when adjusting for elevation, the shooter can turn a knob on the scope, and as she hears
the clicks, she knows that she has adjusted her scope however many minutes of angle.
N ow when she re - aims the crosshairs on the target, the barrel will have a slightly different
angle than it did before, essentially aiming higher to accommodate the longer distance.
To achieve high accuracy in the field, the shooter would know that she zeroed her rifle
to a certain range. Then, when attempting a shot, she would estimate the range to her
target and adjust the scope however many clicks up or down. The biggest cause of error
is an inaccurate estimate of range. Modern shooters often use laser range finders to
determine exactly what elevation offset is required. In long-distance shooting situations,
you can provide the range to the users and let them adjust the rifle’s scope from the
current zero range to a new zero range.
Most games today don’t model even the effect of gravity on a bullet, so adding elevation
adjustment to a sniper or hunting portion of your game will add much-needed accuracy.
Wind
Just like in the Chapter 6 projectile example, our target shooting game’s bullets are
affected by wind. Just as before, a bullet’s susceptibility to wind largely depends on its
lateral drag coefficient. In our simulation, you can tune the bullet’s susceptibility to the
wind by adjusting the factor Cw. This, again, will apply only to rifles shooting at long
ranges. At 20 meters, the wind will have little to do in determining where a bullet will
hit. At 600 meters, it can cause the bullet to be off by a meter! The adjustment for wind
Taking Aim | 359
www.it-ebooks.info
is similar to the adjustment for elevation. By turning a knob, you can adjust your scope
or sights to be slightly to the right or left of the barrel’s centerline. This angle, called
gamma in our simulation, enables you to cancel out the effect of wind.
To deal with wind in the field, we require simple calculations that depend on the par¬
ticular performance of the shooter’s ammunition. It may work something like this: if
the wind is blowing directly across your shot, the adjustment in inches is going to be
half the wind speed in miles per hour. Of course, there are many other rules of thumb
that differ for each caliber of ammunition, but for our simulation you can tune the wind
drag coefficient to whatever value you want your shooter to encounter. He will have to
spend some time at the range, just like a real shooter, figuring out how much the wind
affects his shots. As the wind changes, he will have to adjust in real time and either aim
to the left or right of the target or readjust the windage settings for the scope.
Breathing and Body Position
Although most games don’t model gravity and wind when calculating bullet trajectories,
many do attempt to regulate the accuracy with which you initially fire the bullet. Most
commonly, game developers accomplish this by approximating the crosshairs with four
lines that do not intersect. When fired, the bullet will land anywhere within the circle
described by the inner endpoints of these four lines. Different weapons have different
accuracies, and the lines can move in or out to reflect that. Usually the first shot is the
most accurate, and once the weapon is fired you must “resight” the target, and this takes
time. Therefore, shots fired in quick succession usually become less and less accurate.
In our simulation, we modeled a few things that affect accuracy in the real world and
will give you some suggestions for other factors you could easily include. As our game
was most interested in long-distance shooting via rifles, the most common source of
error is breathing. As discussed before, when a shooter is looking through a scope on a
rifle, she essentially becomes a fixed body. As she breathes, the rifle is essentially breath¬
ing too. When making difficult shots, it is very common for the shooter to take a breath
and hold it while firing. In our simulation, we’ve modeled this with a breathing class
that adjusts the point of aim up and down with time to mimic how a scope moves when
the shooter is breathing deeply. This works via a timer started in our initCompo
nentsQ function that fires every 100 ms. In the code that follows, you’ll see that this
leads to a breath every two seconds.
timer = new Timer(100, TargetPanel);
timer.start();
That function causes the aim point (aimX, aimY) to be moved independently of the
cursor via the following algorithm:
if (direction == true) {
breathHeight = breathHeight + 1;
if (breathHeight == 5) {
360 | Chapter 18: Guns and Explosions
www.it-ebooks.info
direction = false;
breathHeight = breathHeight + 1;
}
}
if (direction == false) {
breathHeight = breathHeight - 1;
if (breathHeight == -5) {
direction = true;
}
}
if (breathing) {
aimY = aimY + breathHeight;
}
Here we are simply moving it up to some limit—in our case, 5 pixels—and then moving
it back down. A better implementation would increase the unsteadiness as the user
zooms in, as shakes are magnified also. There is no limit to the innovative functions you
can use to move the aiming circle away from the cursor to simulate the reality of having
to aim a gun. Yet this is certainly an area of first-person shooters that is lacking in variety.
When he is ready to fire, the user can left-click to hold his breath, the variable breath
ing becomes false, and the crosshairs will stop moving. This simple addition makes
the game much more challenging and engaging. It should be noted that if the shooter
holds his breath too long in real life, the aim will again become unsteady as his body
reacts to not having fresh oxygen. Another improvement to this algorithm would be for
the aim to become unstable after the left mouse button is pressed for some amount of
time.
Many games also change the accuracy of a weapon depending on body position. There
are three basic types of shooting positions: standing, kneeling, and prone. Standing is
—you guessed it—standing up. Kneeling is some form of squatting rather than just
kneeling on your knees. Prone is laid flat on the ground. Because the rifle is locked to
your body, the less unstable your body is, the less unstable the aim. When standing,
your body’s muscles have to do a lot of work to remain upright. When kneeling, they
do less so, and when prone, your muscles don’t have to worry about keeping you standing
at all.
You can add these parameters as random twitches in the aim and tune them to change
the relative advantage of each position. However, prone should always be more stable
than kneeling, and kneeling more stable than standing.
Recoil and Impact
Now that we’ve aimed, fired, and figured out where the bullet is at any given moment,
let’s talk about the last phase, terminal ballistics. To really understand what happens at
Recoil and Impact | 361
www.it-ebooks.info
the end of a bullet’s flight, let’s revisit the beginning. Earlier we talked about recoil as
the result of Newton’s conservation of momentum. Everyone has seen a cheesy movie
where the hero shoots the bad guy and the bullets cause the bad guy to be blown off his
feet. There is a big problem with this! If the bullets were powerful enough to knock the
person they hit off his feet, then the person shooting the gun would also be blown off
his feet! In reality, the force felt by the person being shot is nearly the same as the force
felt when the weapon recoils. For a 9 mm bullet weighing 7.45 g and leaving the barrel
at 390 m/s, the gun will experience recoil such that its momentum is equal to the moment
of the bullet.
One interesting way to incorporate recoil into a video game is in space. On Earth, a gun’s
recoil is pretty quickly transferred to the ground by friction between the player and the
big bad earth. In space, the shooter has no planetary body to push against, so the recoil
of the gun becomes the recoil of the gun/person system. Next time your character needs
to move from one ship to another in a micro-gravity environment, you can make her
spend some ammo to get herself moving.
Now, if you get shot you will probably fall down pretty quickly, but this has more to do
with biology than physics. However, ignoring living targets, if you want to simulate the
damage done by a bullet hitting something, it is more important to look at the bullet’s
kinetic energy. In fact, bullets and artillery shells are called kinetic weapons, as their
primary means of destroying a target is by transferring their kinetic energy to the target.
This is different than, say, a bomb that transfers its chemical energy into heat and kinetic
energy after impact.
Explosions
Accurately modeling explosions involves multiphysics fluid simulations like the kind
discussed in Chapter 14 through Chapter 16. One of our pet peeves is that video games
usually have a collection of barrels lying around that, if shot once, explode violently
enough to blow up nearby vehicles. While this makes for an easy out against multiple
enemies, it is actually pretty hard to get everyday objects to blow up. Shooting a gas can
with a handgun will almost never result in a fire, much less an explosion. Indeed, even
shooting a propane tank with a rifle won’t give you fireworks. It would take something
like a tank of 1/4 propane mixed with 3/4 oxygen to explode, and those aren’t usually
lying around. Regardless, when we play video games we’re often thankful that we have
an occasional red barrel to shoot, so we’ll review how to make the resulting explosion
more accurate even if the ignition is improbable.
362 | Chapter 18: Guns and Explosions
www.it-ebooks.info
Particle Explosions
For most in-game explosions, it will be sufficient to implement a particle-type explosion
that we covered way back in Chapter 2. Now, in Chapter 2 the particles were simply
dots, but they don’t have to be limited to such simple sprites. In some cases, such as
sparks from a bullet hitting a metal container, it would be very accurate to model ex¬
plosions as particles; however, by making our particles look like bits of cars, we can also
make it appear that the car itself exploded. The reason this is easier is because the particle
explosions don’t have any angular motion. Although you can assign different parts of
the cars to different particles, when they fly off due to the explosion, they won’t be
rotating. The good news is that a particle explosion will still give you a realistic distri¬
bution of fragments of something on the ground.
To talk about how to link bullets and particle explosions in detail, we’ll consider some¬
thing more physically accurate than a bullet blowing up a car. Let’s consider a bullet
hitting some loose gravel. This will generally cause the gravel to be thrown up into the
air from the bullet collision. Instead of trying to calculate the complex collisions during
impact, we’ll generate a particle explosion based on the code in Chapter 2:
void CreateParticleExplosion(int x, int y, int Vlnit, int life,
float gravity, float angle)
{
Int l;
Int m;
float f;
Explosion.Active = TRUE;
Explosion.x = x;
Explosion.y = y;
Explosion.V0 = Vlnit;
for(l=0; 1<_MAXPARTICLES; 1++)
{
Explosion.p[l].x = 0;
Explosion.p[l].y = 0;
Explosion.p[l].vl = tb_Rnd(Vlnlt/2, Vlnit);
lf(angle < 999)
{
lf(tb_Rnd(0,l) == 0)
m = -1;
else
n = 1;
Explosion.p[l] .angle = -angle + pi * tb_Rnd(0,10);
} else
Explosion.p[l].angle = tb_Rnd(0,360);
f = (float) tb_Rnd(80, 100) / 100.0f;
Explosions | 363
www.it-ebooks.info
Explosion.p[i].life = tb_Round(llfe * f);
Explosion.p[i]. r = 255;//tb_Rnd(225, 255);
Explosion.p[i].g = 255;//tb_Rnd(85, 115);
Explosion.p[i].b = 255;//tb_Rnd(15, 45);
Explosion.p[i].tine = 0;
Explosion.p[i].Active = TRUE;
Explosion.p[i].gravity = gravity;
}
}
As you can see, the initial velocity V0 controls the strength of the explosion. In Chap¬
ter 2, we chose this value randomly. Now that we have a bullet flying through the air,
we can make a better estimate of how strong of an explosion to create. As you recall
from earlier in the chapter, a bullet has an energy associated with it at any time, t, in its
flight. This energy is its kinetic energy and is equal to half the bullet mass times its
velocity squared.
In our proj ectile simulation, it is simple to calculate this energy as the bullet flies through
the air. It should be noted that a big bullet moving slowly is just as powerful as a smaller
bullet moving quickly. Our upcoming code is going to assume that 100% of the kinetic
energy is delivered to the target. This would not be true if a bullet shot straight through
something. A way to visualize this is to imagine two targets, both hanging from the
ceiling. One is made from paper and one is made from steel. When shot at, the steel
target swings from its support, while the paper target stays still. This is because the bullet
is passing straight though the paper and not transferring its kinetic energy to the target.
To make things simple, we’ll transfer all of the bullet’s kinetic energy to the gravel. In
equation form, this would look like:
1/2 mb v“bullet — 211/2 mg v"g rave j
Note that this is the sum of the individual bits of gravel’s velocities. In, Chapter 2, each
particle was just given a random velocity anywhere from Vinit/2 to Vinit. This could
lead to creating a set of particles whose energies exceed the input energy. To prevent
this, we’ll add a variable to our explosion class like so:
typedef struct _TPartlcle
{
float
float
float
float
int
x;
y;
vt;
angle;
life;
// x-coordlnate of the particle
// y-coordlnate of the particle
// initial velocity
// initial trajectory (direction)
// duration in milliseconds
// red component of particle's color
// green component of particle's color
// blue component of particle's color
// keeps track of the effect's time
// gravity factor
int
int
int
int
r;
g;
b;
time;
gravity;
float
364 | Chapter 18: Guns and Explosions
www.it-ebooks.info
BOOL
Active;
float mass;
} TParticle;
#define _MAXPARTICLES 50
#define _MASSOFPARTICLE .25
// indicates whether this particle
// is active or dead
//for calculating the particle's energy
typedef struct _TParticleExplosion
{
TParticle
p[_MAXPARTICLES]; // list of particles
// making
up this effect
int
x;
// initial x location
int
y;
// initial y location
float
float
KE;
//Available kinect energy
BOOL
Active; // indicates whether
this effect is
//active or dead
} TParticleExplosion;
Notice that V0 is no longer required, as the available kinetic energy will govern the
strength of the explosion. Assuming that the bullet’s kinetic energy is given as a variable
KEb, our new CreateParticleExplosion function would look like the following:
void CreateParticleExplosion(int x, int y, int KEb, int life,
float gravity, float angle)
{
int i;
int m;
float f;
Explosion.Active = TRUE;
Explosion.x = x;
Explosion.y = y;
Explosion.KE = KEb;
for(i=0; i<_MAXPARTICLES; i++)
{
Explosion.p[i].x = 0;
Explosion.p[i].y = 0;
Explosion.p[i].m = _MASSOFPARTICLE; //Mass of a single gravel
Explosion.p[i].vt = tb_Rnd(0, sqrt(Explosion.KE/(_MASSOFPARTICLE*
_MAXPARTICLES));
Explosion.KE = Explosion.KE - ((l/2)*(Explosion.p[i].m)*
(Explosion.p[i].vi));
if(angle < 999)
{
if(tb_Rnd(0,1) == 0)
m = -1;
else
n = 1;
Explosions | 365
www.it-ebooks.info
Explosion.p[1].angle = -angle + m * tb_Rnd(0,10);
} else
Explosion.p[1].angle = tb_Rnd(0,360);
f = (float) tb_Rnd(80, 100) / 100.0f;
Explosion.p[i].life = tb_Round(life * f);
Explosion.p[1].r = 255;//tb_Rnd(225, 255);
Explosion.p[i].g = 255;//tb_Rnd(85, 115);
Explosion.p[i].b = 255;//tb_Rnd(15, 45);
Explosion.p[i].tine = 0;
Explosion.p[i].Active = TRUE;
Explosion.p[i].gravity = gravity;
}
As you can see, we’ve altered the statements that set the initial velocity of the particles
to be a random-number generator in a range anywhere from 0 to a velocity that would
consume the entire explosions kinetic energy. The next line reduces the available kinetic
energy in the explosion by the amount just assigned to the particle. This way, you can
be sure that the outgoing explosion is never more powerful then the input. A more
interesting way to handle this would be to first initialize the particles with some given
mass distribution and to assign the velocities not randomly, but with a normal distri¬
bution. Numerical recipes in C can help you accomplish this.
Even though the preceding code does not take into account some of the more subtle
aspects of the transfer of kinetic energy, it will ensure that a small, slow-moving bullet
produces a smaller explosion than a big, fast-moving one. This is something that is
lacking in todays video games.
Polygon Explosions
While particle explosions are appropriate for small, uniform objects, they fail to give
appropriate realism when something is blown into identifiable chunks. This is why in
video games you rarely see a car explode and the door fly away to land next to you.
Instead, games usually handle objects like this with a particle explosion that obscures
the object while it is re-rendered in its now-exploded state with the missing pieces having
been apparently blown to smithereens.
If you do want to model a full explosion of solid bodies, you can reuse the particle code
for the translation aspects. Essentially the particles will now describe the center of gravity
of each solid body. You will have to add in an initial angular velocity and let the simu¬
lation, as described in Chapter 12, handle their motion after that initial angle.
While we don’t have room to go over another example here, we’ll talk a little about the
input energy to such an explosion to help you bridge the gap. While we are on that
subject, let’s recall that a bullet just doesn’t have the energy required to blow something
apart. Even when you hit something with a tank-mounted gun, it really isn’t the kinetic
366 | Chapter 18: Guns and Explosions
www.it-ebooks.info
energy of the bullet that blows apart the thing you hit, but some secondary explosion.
In the case of a tank hitting another tank, the molten slag from the impact is usually
peppered all over the inside of the tank, causing the fuel or ammunition to explode.
That is where you get the big booms—it’s the conversion of chemical energy to heat,
light, and pressure!
The most common method of quantifying the chemical energy in weapons is called
TNT equivalency. This is how much TNT it would take to cause the same explosion
regardless of what you are actually exploding. Now, explosion modeling of, say, gasoline
and air is pretty complex, so let’s stick with TNT. A kilogram of TNT contains 4.184
Mega joules of energy; a 9 mm round has 400 J. You can see from that comparison why
it is hard to blow something up by shooting at it, but easy to do with a block of TNT.
For the purposes of this discussion, let’s say you have an open box (five polygon sides)
into which your player just tossed a 1 kg block of TNT. When the TNT is detonated,
you can give each polygon side an initial velocity (translational and angular) and let the
kinematic equations take over. Those velocities can be based on two simple rules.
• The velocity vector can be defined by two points: the center of the block of TNT
and the center of area of the polygon.
• The sum of all the kinetic energy must be less than the available chemical energy
in the TNT. This can be prorated by the square of the distance from the polygon to
the block of TNT.
The use of the center of area in our first rule will impart some rotation into our polygon,
as it will cause a moment about the center of gravity unless the two coincide. If this is
the case, as it would be for our box, aerodynamic drag and unevenness of the explosion
will still cause rotation, so you should either model these explicitly or impart some
rotational velocity manually.
Now that we have a velocity direction, we need to define its magnitude. The force on
objects near an explosion is caused by the rapid expansion of gasses due to the heat
generated by the detonation of the explosive. However, not all the chemical energy is
transferred to the objects—a lot of it is converted into heat, light, and sound. Typically,
only one-third of the available chemical energy is converted in the initial detonation.
Let’s call this the efficiency of the explosion, which we’ll denote by (. Therefore, we can
write the relationship between velocities of the polygons as follows:
CE chemical = 2 y(m,.v , 2 + W)
i=0 ^
Here, we can tune (to give the polygons realistic velocities in the event of an explosion
(i.e., not sending them off at the speed of light). If the energy from the explosion is
Explosions | 367
www.it-ebooks.info
divided up equally, you can see that lighter objects will have higher velocities, as you
might expect. You can also adjust the amount of explosion energy available to each
object by weighting the objects imparted energy by its distance from the explosion. This
should also be tuned in the program, but in general, pressure from an explosion de¬
creases with the cube of distance and exponentially with time.
If you want to model more complex explosion-structure interactions, there are many
good references for how, say, buildings, react to bomb blasts. FEMA, as well as the Army
and Navy, have several papers on the subject, such as FEMA 426 Reference Manual to
Mitigate Potential Terrorist Attacks Against Buildings. The general concepts laid out in
such documents can increase the realism of the building damage due to explosions.
368 | Chapter 18: Guns and Explosions
www.it-ebooks.info
CHAPTER 19
Sports
The topic of sports is nearly as vast as all of the subjects we’ve covered combined. There
is a sport for everyone, and a sport that takes advantage of each of the physical models
we’ve discussed so far. The topic ranges from games full of accessories, such as golf or
polo, to running, where all you need are your own two feet.
One of the most attractive aspects of sports for the game programmer is that they take
place in a limited physical space by design. Unlike a first-person shooter where the player
will eventually reach an artificial boundary, in a sports game the player will not expect
to be able to walk out of the court. Almost all sports have defined dimensions that are
relatively easy to model. Table 19-1 lists a few sports and their professional field di¬
mensions.
Table 19-1. Various field dimensions
1 Sport
Field size J
Soccer (football)
90-120 m long by 45-90 m wide
Football (including end zones)
109.7 m long by 48.8 m wide
Baseball
27.4 m between bases; 18.39 from pitcher's mound to home base; outfield varies
Basketball (international)
28 m by 15 m
Ice hockey (international)
61 m by 30 m
As you can see, other than baseball—where the shape of the outfield changes depending
on what stadium you are in—modeling these field sizes is a rather straightforward ex¬
ercise.
Additionally, the one thing that sports have in common is that they have a human actor.
In this chapter we’ll explore how the human action can be simulated as input for the
other physical simulations we’ve discussed. Specifically, we’ll show you an example of
how to model a person swinging a golf club using accurate physiological models. This
is called biomechanics. Before we get into that, another important thing to understand
369
www.it-ebooks.info
when you’re modeling sports is the limits of the human body. Although records are
broken in every Olympics, no human being is able to jump 10 feet vertically into the
air. Unless you are breaking the limits of biomechanics on purpose, doing so will de¬
crease the realism of your game. The biomechanical statistics of what would be con¬
sidered an outstanding athlete are given in Table 19-2.
Table 19-2. Table of human performance
1 Physical attribute
Average value
Record value 1
Jump from standstill (vertical)
81 cm
155 cm
Running jump height (high jump)
1.83 m
2.45 m
Jump distance
5.0 m
8.95 m
Throwing speed
24.5 m/s
46.0 m/s
Running speed over 100 m
7.5 m/s
10 m/s
Running speed over 10,000 m
3.7 m/s
6.3 m/s
Almost all sports records are available online somewhere, so Table 19-2 is by no means
exhaustive. However, it is a good idea to use these values to limit your simulations of
human actions in your video games. Obviously, part of the excitement of playing video
games is to be able to jump higher and run faster than you otherwise could, but a good
survey of biomechanics will at least let you know what is extraordinary and what is not.
Now let’s take a look at how we would model a human actor in a sports game.
Modeling a Golf Swing
Let’s say you’re writing a golf game and you want include a little realism. An obvious
important element of the game is the golf swing. Another is the club-to-ball impact, and
still another is the trajectory of the ball in flight. You can use the projectile motion
modeling techniques discussed earlier in Chapter 2, Chapter 4, and Chapter 6 to model
the ball’s flight, and the collision response techniques in Chapter 5 to model the club-
to-ball impact. But what about the golf swing?
Well, before we show you one way to model a golf swing, let’s talk about why you would
want to do so in the first place. To model club-to-ball impact, you need to know the club
head velocity at the time of impact. That velocity is a function of the swing. The golfer
raises the club through his backswing, torques his body, and brings the club head down
in an arc while applying a torque with his wrists. As the club swings down, the wrist
torque reverses, and the club whips through the downswing until the club head collides
with the ball. (Or, in our case, collides with the ground!) Now, there are many subtle
details we’ve omitted here with regard to technique and the physics, but you get the
idea. At any rate, the swing determines the club head velocity at the moment of impact,
which in turn determines the velocity of the ball after impact.
370 | Chapter 19: Sports
www.it-ebooks.info
If you were writing a game for the Wii or some other platform that can capture a player’s
motion, then you can relate the player’s swing motion to the initial torque applied to a
virtual golf club, thus determining, through some model, the swing dynamics and re¬
sulting club head velocity.
Golfers take swing technique seriously and so do scientists who study golf swing dy¬
namics. In an effort to understand what makes a good swing or how to improve a swing,
there are many scientists out there actively studying the golf swing physics. As a result,
there are many mathematical models of varying degrees of realism and complexity that
aim to examine the golf swing. One example is the so-called two-rod model as described
in Theodore P. Jorgensen’s book The Physics of Golf. In his book, Dr. Jorgensen describes
the two-rod model in detail, including assumptions and simplifications, and provides
the resulting equations that must be solved to simulate a golf swing based on this model.
He even provides empirical data used to validate the results of the mathematical model.
As shown in Figure 19-1, the two-rod model assumes that the golfer’s arm is one rod
that extends from the shoulders to the wrists. This is the arm rod. The club is represented
by another rod that extends from the wrist end of the arm rod to the club head.
This model is essentially a double pendulum. More specifically, it is a driven double
pendulum since the model assumes a torque applied at the shoulder end of the arm rod,
and another torque applied at the wrist joint connecting the arm rod to the club rod.
Modeling a Golf Swing | 371
www.it-ebooks.info
We won’t repeat Dr. Jorgensen’s development of the model here; instead, we’ll show you
how to solve the resulting equations:
Those equations can be found in the Technical Appendix—Section 4 of Dr. Jorgensen’s
book—but are listed here for convenience.
Equation 1:
(J+I+M c R 2 +2RScos(p))a-(I+RS-cos(p))'p+
+( P 2 -2ap)RSsin( p)-S[gsin( 9+fi)-a-cos( 6+P) ]-
-(S A +M c R)(g-sin(9)-a-cos(9))=Q a
Equation 2:
iP-[I+RS-cos(P)]a+a 2 RS-sin(P)+S[g-sin(6+P)-a-cos(6+P)]=Q p
Table 19-3 explains what each symbol represents.
Table 19-3. Symbols used in golf swing model
Symbol Meaning
J Mass moment of inertia of the rod representing the arm. Units are kg-m 2 .
I Mass moment of inertia of the rod representing the dub. Units are kg-m 2 .
M c Mass of the club. Units are kg.
R Length of the rod representing the arm. Units are m.
S First moment of the rod representing the club about the wrist axis (where the club rod connects to the arm rod). Units
are kg-m.
a Angle swept by arm rod from initial backswing position. Units are radians.
(5 Wrist-cock angle. Units are radians,
g Acceleration due to gravity. Constant 9.8 m/s 2 .
0 Angle between arm rod and vertical axis. Units are radians,
a Horizontal acceleration of the shoulder. Units are m/s 2 .
S A First moment of the arm rod about the shoulder axis. Units are kg-m.
Q„ Torgue applied at the shoulder to the arm rod. Units are N-m.
Qp Torgue applied at the wrist joint to the club rod. Units are N-m.
These equations represent a coupled system of nonlinear differential equations. They
are coupled in that they both depend on the unknown quantities a and (J. They are
clearly differential equations, as they both include time derivatives of the unknown
372 | Chapter 19: Sports
www.it-ebooks.info
quantities. And they are nonlinear because they include sines and cosines of one of the
unknowns along with derivatives of the other unknown raised to some power greater
than 1.
So, how do we solve these equations? Well, we can’t do so in closed form and must resort
to numerical means. There are a number of ways to proceed, but the approach we’ll use
is to first solve Equation 2 for [3 and substitute the result into Equation 1. Then, we’ll
numerically integrate the result using a fourth-order Runge-Kutta scheme, as described
in Chapter 7.
More specifically, at each time step Equation 1 with j3 replaced by the expression derived
from Equation 1 will be solved for a. Once a is found, we can find ft using the second
equation previously solved for (3. Next, we can integrate a and [3 to find a and (3. This
process then repeats for each time step.
Again, this is only one method of solving these equations. Normally, when faced with
a system of equations, practitioners use matrix schemes to solve the equations simul¬
taneously. This is almost necessary for systems of equations that involve more than two
equations. However, with just two equations, as we have here, we can avoid expensive
matrix inversion computation by using the technique we just described.
Solving the Golf Swing Equations
Now we’ll show you how to implement the solution we described in a simple console
application. The example solves the two governing equations for a and (3 over time, the
results of which can then be used to determine the club head velocity at any time instant
using kinematic equations as described in Chapter 2 (see the section “Rigid-Body Kin¬
ematics” on page 61). Alternatively, you can use the following equation, which Dr. Jor¬
gensen gives for the club head velocity in his book:
V 2 =[« 2 +Lr +2itZ’Cos(/?)]cr +L 2 j3 2 - 2 [z . 2 -i- AL ■ cos(/3 )]«>0
We’ll use Jorgensen’s equation in this example.
Since the angles of interest are computed in units of radians, but we want to report them
in units of degrees, we first create a few defines to make the conversions for us:
#define RADIANS(d) (d/180.0*3.14159)
#define DEGREES(r) (r*180.0/3.14159)
Next, we declare and initialize all of the variables. The initial values used here are some
typical values that we assumed. You can change these values to simulate different swings:
// Variables
double alpha = 0.0;
double alpha_dot = 0.0;
double alpha_dotdot = 0.0;
double beta = RADIANS(120.0);
Modeling a Golf Swing | 373
www.it-ebooks.info
double beta_dot = 0.0;
double beta_dotdot =0.0;
double 3 = 1.15; //kg m A 2
double I = 0.08; //kg m A 2
double Me = 0.4; // kg
double R = 0.62; // n
double L = 1.1; // m
double S = 0.4*1.1*0.75; // kg m
double g = 9.8; // m/s A 2
double gamma = RADIANS(135.0);
double theta = gamma - alpha;
double SA = 7.3*0.62*0.5; // kg m
double Qalpha = 100; // N m
double Qbeta = -10; // N m
double a = 0.1*g; // m/s A 2
double dt = 0.0025; // s
double time =0; // s
double Vc = 0;
Next we define two functions that we will use to compute the second time derivatives
of a and (3 (i.e., a and ji ). These functions simply use Equations 1 and 2 solved for a and
jS, respectively.
ComputeAlphaDotDot, which solves for a, is shown here:
double ComputeAlphaDotDot(vold)
{
double A, B, C, D, F, G;
double num, denom;
A=(J+I+Mc*R*R+2*R*S* cos(beta));
B = -(I + R * S * cos(beta));
F = Qalpha - (beta_dot * beta_dot - 2 * alpha_dot * beta_dot) * R * S *
sin(beta) + S * (g * sin(theta + beta) - a * cos(theta + beta))
+ (SA + Me * R) * (g * sln(theta) - a * cos(theta));
C = B;
D = I;
G = Qbeta - alpha_dot * alpha_dot * R * S * sln(beta) -
S * (g * sin(theta + beta) - a * cos(theta + beta));
num = (F - (B * G / D));
denom = (A-(B*C/D));
return (F - (B * G / D)) / (A-(B*C/D));
}
The local variables A, B, C, D, F, and G are convenience variables used to organize the
terms in Equation 1. This function returns the second derivative of a.
ComputeBetaDotDot, shown next, is very similar to ComputeAlphaDotDot but solves
Equation 2 instead. This function returns the second derivative of (3:
374 | Chapter 19: Sports
www.it-ebooks.info
double ComputeBetaDotDot(vold)
{
double C, D, G;
C = -(I + R * S * cos(beta));
D = I;
G = Qbeta - alpha_dot * alpha_dot * R * S * sln(beta) -
S * (g * sin(theta + beta) - a * cos(theta + beta));
return (G - C * alpha_dotdot) / D;
}
The solution to Equations 1 and 2 follows the Runge-Kutta scheme we showed you in
Chapter 7. Four intermediate steps are taken for each time step. The time step size is
controlled by dt, which we’ve set to 0.0025s. If you simply used Euler’s method, you’d
have to reduce this step size quite a bit to obtain a stable solution. We implemented the
solution in the main function of our console example. The code is as follows:
Int _tnaln(lnt argc, _TCHAR* argv[])
{
double a, at;
double b, bt;
int 1;
FILE* fp;
double phi;
double Vc2;
double akl, ak2, ak3, ak4;
double bkl, bk2, bk3, bk4;
FILE* fdebug;
fp = fopen("results.txt", "w");
fdebug = fopen("debug.txt", "w");
for(l = 0; i<200; 1++)
{
tine += dt;
if(tine>=0.1)
{
Qbeta = 0;
}
// save results of previous tine step
a = alpha;
b = beta;
at = alpha_dot;
bt = beta_dot;
// integrate alpha'' and beta 1 '
Modeling a Golf Swing | 375
www.it-ebooks.info
// The K1 Step:
alpha_dotdot = ComputeAlphaDotDot();
beta_dotdot = ComputeBetaDotDot();
akl = alpha_dotdot * dt;
bkl = beta_dotdot * dt;
alpha_dot = at + akl/2;
beta_dot = bt + bkl/2;
// The K2 Step:
alpha_dotdot = ComputeAlphaDotDot();
beta_dotdot = ComputeBetaDotDot();
ak2 = alpha_dotdot * dt;
bk2 = beta_dotdot * dt;
alpha_dot = at + ak2/2;
beta_dot = bt + bk2/2;
// The K3 Step:
alpha_dotdot = ComputeAlphaDotDot();
beta_dotdot = ComputeBetaDotDotQ;
ak3 = alpha_dotdot * dt;
bk3 = beta_dotdot * dt;
alpha_dot = at + ak3;
beta_dot = bt + bk3;
// The K3 Step:
alpha_dotdot = ComputeAlphaDotDot();
beta_dotdot = ConputeBetaDotDot();
ak4 = alpha_dotdot * dt;
bk4 = beta_dotdot * dt;
alpha_dot = at + (akl + 2*ak2 + 2*ak3 + ak4) / 6;
beta_dot = bt + (bkl + 2*bk2 + 2*bk3 + bk4) / 6;
alpha = a + alpha_dot * dt;
beta = b + beta_dot * dt;
theta = gamma - alpha;
Vc2 = (R*R + L*L + 2 * R * L * cos(beta)) * ( alpha_dot * alpha_dot)
+ L*L * beta_dot * beta_dot
- 2 * (L*L + R * L * cos(beta)) * alpha_dot * beta_dot;
Vc = sqrt(Vc2);
phi = theta + beta;
fprintf(fp, "%f, %f, %f, %f, %f, %f\n", time, DECREES(theta),
376 | Chapter 19: Sports
www.it-ebooks.info
DEGREES(alpha), DEGREES(beta), DEGREES(phi), Vc);
fprintf(fdebug, "%f, %f, %f, %f, %f, %f , %f\n", tine, DEGREES(alpha),
alpha_dot, alpha_dotdot, DEGREES(beta), beta_dot, beta_dotdot);
}
fclose(fp);
fclose(fdebug);
return 0;
}
Local variables a, at, b, and bt are used to temporarily store the previous time step’s
results for a and (3 and their first derivatives, i is a counter variable, f p is a file pointer
that we’ll use to write results out to a text file, phi is used to store the sum of 0 + (3. And
Vc2 is the square of the club head velocity calculated according to Jorgensen’s equation.
The variables akl through ak4, and bkl through bk4, are used to store intermediate
results of the Runge-Kutta integration scheme, f debug is a file pointer to a file we used
for writing debugging information.
After the output and debug files are opened, the function enters a loop to perform the
integration over 200 time steps. You can change the number of time steps as you see fit
for your application. Keep in mind that the swing event, from start to striking the ball,
takes place over a very short period of time—only fractions of a second long.
Upon entering the loop, you’ll see some code that checks how much time has elapsed;
if that time is greater than 0.1s, the wrist torque, Qbeta, is set to 0. This is a crude model
of how the wrist torque that’s initially applied is released, allowing the club to swing past
the arm. Depending on the swing you’re modeling, this torque could actually reverse
direction, forcing the club past the arm even more. Dr. Jorgensen’s book explains all this
in detail, even giving experimental results.
Next, results of the previous time step are saved in the variables a, b, at, and bt. The
first time step simply stores the initial values. Now, the integration starts for the first
step, kl (see Chapter 7). Each of these steps involves computing a and ft using the
functions ComputeAlphaDotDot and ComputeBetaDotDot. The kl results are then cal¬
culated and used to compute intermediate results for the first time derivatives of a and
(3. All four intermediate steps are carried out in a similar manner.
Finally, the current time step’s results for alpha_dot and beta_dot, along with alpha
and beta, are computed. Also, the square of the club head velocity, Vc2, is computed
using Jorgensen’s equation shown earlier; and the club head velocity, Vc, results from
the square root of Vc2.
Modeling a Golf Swing | 377
www.it-ebooks.info
Results of interest are then written to the output and debug files. So, that’s pretty much
it. After the loop finishes, the files are closed and the application terminates.
If this were an actual game, you would use the club head velocity results along with the
collision response method we showed in Chapter 3 to determine the golfball’s trajectory.
You could model the flight path of the golf ball using the methods we showed you in
Chapter 6.
Billiards
Now let’s take a look at a different example. You may not think of billiards as a sport,
but it is recognized internationally as a cue sport. Cue sports are a family of sports that
include billiards, pool, snooker, and other related variations. For simplicity, we’ll stick
with the term billiards, although the topics presented apply to all cue sports.
Billiards is a good example of an activity that takes place over a limited physical space.
Thus, when writing a billiards video game you need only concern yourself with a very
finite space composed of well-established geometry. Billiard tables are typically 1.37 m
x 2.74 m (4.5 ft x 9 ft), with some longer and some smaller depending on the game,
style, and space available. Tables are typically cloth-covered slate. Balls vary in size be¬
tween games and regions, with American-style pool balls measuring about 57 mm (2.25
inches) in diameter. Balls used to be made of wood, clay, or ivory, but nowadays they
are plastic.
All these characteristics are important little details that you must consider if you’re going
to make a realistic billiards video game. The slate table and hard plastic balls have certain
impact characteristics. The cloth-covered table provides some resistance to rolling. Side
bumpers are not as hard as the slate table, thus yielding different impact characteristics.
Fortunately, data on billiard tables and balls is readily available on the Web. And sim¬
ulating billiards in a video game is fairly straightforward.
Billiards makes an interesting example because collisions are the heart of the game, and
such an example also gives us an opportunity to demonstrate rolling contact. Figure 19-2
and Figure 19-3 illustrate the example we’ll focus on. We have three object balls (the
ones that get struck with the cue ball) set up in the middle of the table in a loose triangle
configuration.
378 | Chapter 19: Sports
www.it-ebooks.info
Figure 19-2. Elapsed time = 0.398s
The cue ball comes from the right at a set speed (see Figure 19-2) and then impacts the
eight ball (see Figure 19-3).
Figure 19-3. Elapsed time = 0.440s
Billiards | 379
www.it-ebooks.info
After the initial impact between the cue ball and the eight ball, the eight ball moves to
the left and impacts two more balls. These balls then shoot off diagonally. Most of the
energy from the eight ball is transferred to the two other balls, so the eight ball quickly
comes to rest while being kissed by the cue ball. The other two balls continue rolling
away diagonally (see Figure 19-4).
Figure 19-4. Elapsed time = 0.566s
In this example, we’ll show you how to handle ball-ball collisions, ball-table collisions,
ball-table contact, aerodynamic drag on the ball, rolling resistance, friction between
balls at the time of impact, and friction between the balls and table.
Implementation
If you’ve read and studied the examples presented in Chapter 7 through Chapter 13,
then the implementation of this billiards example will be very familiar to you; we use
the same basic approach. During each simulation time step, we calculate all the forces
acting on each ball; integrate the equations of motion, updating each ball’s position and
velocity; and then check for and deal with collisions.
380 | Chapter 19: Sports
www.it-ebooks.info
The rigid-body class used in this example is very similar to that used for the airplane
example in Chapter 15. Even though the balls are compact and round, and it’s tempting
to treat them as particles, you must treat them as 3D rigid bodies in order to capture
rolling and spinning, which are important elements of billiard ball dynamics. The rigid-
body class adopted for this billiards example is as follows.
typedef struct _RtgidBody {
float fMass; // Total mass (constant)
Matrlx3x3 mlnertia; // Mass moment of inertia in body coordinates
Matrix3x3 mlnertialnverse;// Inverse of mass moment of inertia matrix
Vector vPosition; // Position in earth coordinates
Vector vVelocity; // Velocity in earth coordinates
Vector vVelocityBody; // Velocity in body coordinates
Vector vAcceleration; // Acceleration of eg in earth space
Vector vAngularAcceleration; //Angular acceleration in body coordinates
Vector vAngularAccelerationGlobal; // Angular acceleration
// in global coordinates
Vector vAngularVelocity; // Angular velocity in body coordinates
Vector vAngularVelocityGlobal; // Angular velocity in global coordinates
Vector vEulerAngles; // Euler angles in body coordinates
float fSpeed; // Speed (magnitude of the velocity)
Quaternion qOrientation; // Orientation in earth coordinates
Vector vForces; // Total force on body
Vector vMoments; // Total moment (torque) on body
Matrix3x3 mlelnverse; // Inverse of moment of inertia in earth coordinates
float fRadius; // Ball radius
} RigidBody, *pRigidBody;
As you can see, this class looks very similar to the rigid-body classes we’ve used through¬
out this book, and in particular that used in the airplane example. All the usual suspects
are here, and the comments in this code sample state what each class member represents.
One particular property you have not seen yet is fRadius —this is simply the billiard
ball’s radius, which is used when we are checking for collisions and calculating drag
forces.
As we discussed in Chapter 14, since there are multiple objects in this simulation that
may collide, we’re going to iterate through all the objects, checking for collisions while
storing the collision data. Since there are not that many objects in this simulation, we
don’t really need to partition the game space in order to optimize the collision detection
checks (refer to Chapter 14). The data we need to store for each collision is included in
the following Collision structure:
typedef struct _Collision {
Int
bodyl;
int
body2;
Vector
vCollisionNormal;
Vector
vCollisionPoint;
Vector
vRelatlveVeloctty;
Vector
vRelatlveAcceleration;
Billiards | 381
www.it-ebooks.info
Vector vCollislonTangent;
} Collision, *pCollision;
The first two properties are indices to the two bodies involved in the collision. The next
property, vCollisionNormal, stores the normal vector at the point of contact of the
collision with the vector pointing outward from body2. The next property, vCollision
Point, stores the coordinates of the point of contact in global coordinates. Since we’re
dealing with spheres (billiard balls), the collision manifold will always consist of a single
point for each ball-ball or ball-table collision. The next two properties store the relative
velocity and acceleration between the two bodies at the point of collision. The data is
stored in vRelativeVelocity and vRelativeAcceleration, respectively. To capture
friction at the point of contact, we need to know the tangent vector to the bodies at the
point of contact. This tangent is stored in vCollisionTangent.
We set up several global defines to hold key data, allowing us to easily tune the simu¬
lation:
#define
BALLDIAMETER
0.05715f
#define
BALLWEIGHT
1.612f
#define
GRAVITY
-9.87f
#define
LINEARDRAGCOEFFICIENT
0.5f
#define
ANGULARDRAGCOEFFICIENT
0.05f
#define
FRICTIONFACTOR
0.5f
#define
COEFFICIENTOFRESTITUTION
0.8f
#define
COEFFICIENTOFRESTITUTIONGROUND 0.1f
#define
FRICTIONCOEFFICIENTBALLS
0.1f
#define
FRICTIONCOEFFICIENTGROUND
0.1f
#define
ROLLINGRESISTANCECOEFFICIENT 0.025f
The first three defines represent the billiard ball diameter in meters, the ball weight in
newtons, and the acceleration due to gravity in m/s 2 . The ball diameter and weight are
typical values for American-style billiard balls (i.e., 2.25 inches and 5.8 oz on average).
The remaining defines are self-explanatory and represent nondimensional coefficients
such as drag coefficients and coefficients of restitution. The values you see are what we
came up with after tuning the simulation. You’ll surely tune these yourself if you develop
your own billiards game.
We use three important global variables for this simulation, as shown here:
RigidBody Bodies[NUMBODIES];
Collision Collisions[NUMB0DIES*8];
int NumCollisions = 0;
Bodies is an array of RigidBody types and represents the collection of the billiard balls.
Here we’ve defined NUMBODIES as 4, so there are four billiard balls in this simulation.
We’ve adopted the convention that the cue ball will always be Bodies [0].
382 | Chapter 19: Sports
www.it-ebooks.info
Initialization
At the beginning of the simulation, we have to initialize all four billiard balls. We use
one function, InitializeObjects, for this task. It’s a long function, but it’s really simple.
The code is shown on this and the following pages. Bodies [0] is the cue ball, and it is
positioned 50 ball diameters along the negative x-axis away from the object balls. There’s
no magic to this number; we picked it arbitrarily. Now, we did deliberately set the z-
position of the cue ball (all the balls, for that matter) to one-half the diameter so that
the balls would be just touching the table at the start of the simulation.
To have the cue ball roll from right to left, we gave it an initial velocity of 7 m/s along
the positive x-axis. With this initial velocity, the cue ball will begin sliding across the
table for some short distance as it also starts to roll due to the friction between the ball
and table. You can see in the upcoming code sample that all the other kinematic prop¬
erties are set to 0 for the cue ball. For the object balls, all of their kinematic properties
are set to 0.
We encourage you to experiment with different initial values for the cue ball’s kinematic
properties. For example, try setting the angular velocity about any of the coordinate
axes to something other than 0. Doing so will allow you to see how a spinning ball may
move slightly left or right depending on the spin. It’s also fun to see how spin affects the
object balls upon collision with the cue ball.
Aside from setting the positions and kinematic properties of the balls, InitializeOb
jects also initializes mass properties. We used the previously defined BALLWEIGHT
divided by the acceleration due to gravity to determine the ball mass. For mass moment
of inertia, we simply used the equations for a solid sphere we showed you way back in
Chapter 1:
void InitializeObjects(int configuration)
{
float iRoll, iPitch, iYaw;
int i;
float Ixx, Iyy, Izz;
float s;
/////////////////////////////////////////////////////
// Initialize the cue ball:
// Set initial position
Bodies[0].vPosition.x = -BALLDIAMETER*50.0f;
Bodies[0].vPosition.y = 0.0f;
Bodies[0].vPosition.z = BALLDIAMETER/2.0f;
// Set initial velocity
s = 7.0;
Bodies[0].vVelocity.x = s;
Bodies[0].vVelocity.y = 0.0f;
Bodies[0].vVelocity.z = 0.0f;
Bodies[0].fSpeed = s;
Billiards | 383
www.it-ebooks.info
// Set initial angular velocity
Bodies[0].vAngularVelocity.x = 0.0f; // rotate about long'l axis
Bodies[0].vAngularVelocity.y = 0.0f; // rotate about transverse axis
Bodies[0].vAngularVelocity.z = 0.0f; // rotate about vertical axis
Bodies[0].vAngularAcceleration.x = 0.0f;
Bodies[0].vAngularAcceleration.y = 0.0f;
Bodies[0].vAngularAcceleration.z = 0.0f;
Bodies[0].vAcceleration.x = 0.0f;
Bodies[0].vAcceleration.y = 0.0f;
Bodies[0].vAcceleration.z = 0.0f;
// Set the initial forces and moments
Bodies[0].vForces.x = 0.0f;
Bodies[0].vForces.y = 0.0f;
Bodies[0].vForces.z = 0.0f;
Bodies[0].vMoments.x = 0.0f;
Bodies[0].vMoments.y = 0.0f;
Bodies[0].vMoments.z = 0.0f;
// Zero the velocity in body space coordinates
Bodies[0].vVelocityBody.x = 0.0f;
Bodies[0].vVelocityBody.y = 0.0f;
Bodies[0].vVelocityBody.z = 0.0f;
// Set the initial orientation
iRoll = 0.0f;
iPitch = 0.0f;
iYaw = 0.0f;
Bodies[0].qOrientation = MakeQFromEulerAngles(iRoll, iPitch, iYaw);
// Set the mass properties
Bodies[0].fMass = BALLWEIGHT/(-g);
Ixx = 2.0f * Bodies[0].fMass / 5.0f * (BALLDIAMETER/2*BALLDIAMETER/2);
Izz = Iyy = Ixx;
Bodies[0].mlnertia.ell = Ixx;
Bodies[0],mlnertia.el2 = 0;
Bodies[0],mlnertia.el3 = 0;
Bodies[0],mlnertia.e21 = 0;
Bodies[0],mlnertia.e22 = Iyy;
Bodies[0],mlnertia.e23 = 0;
Bodies[0],mlnertia.e31 = 0;
Bodies[0],mlnertia.e32 = 0;
Bodies[0],mlnertia.e33 = Izz;
Bodies[0].mlnertialnverse = Bodies[0].mlnertia.lnverse();
384 | Chapter 19: Sports
www.it-ebooks.info
Bodies[0] .fRadlus = BALLDIAMETER/2;
/////////////////////////////////////////////////////
// Initialize the other balls
for(i=l; i<NUMBODIES; i++)
{
// Set initial position
if(i==l)
{
Bodies[i].vPosition.x
Bodies[i].vPosition.y
Bodies[i].vPosition.z
} else if(i==2) {
Bodies[i].vPosition.x
Bodies[i].vPosition.y
Bodies[i].vPosition.z
} else {
Bodies[i].vPosition.x
Bodies[i].vPosition.y
Bodies[i].vPosition.z
0 . 0 ;
-(BALLDIAMETER/2.0f+0.25*BALLDIAMETER);
BALLDIAMETER/2.0f;
0 . 0 ;
BALLDIAMETER/2,0f+0.25*BALLDIAMETER;
BALLDIAMETER/2.0f;
-BALLDIAMETER;
0.0f;
BALLDIAMETER/2.0f;
}
// Set initial velocity
Bodies[i].vVelocity.x = 0.0f;
Bodies[i].vVelocity.y = 0.0f;
Bodies[i].vVelocity.z = 0.0f;
Bodies[i].fSpeed = 0.0f;
// Set initial angular velocity
Bodies[i].vAngularVelocity.x = 0.0f;
Bodies[i].vAngularVelocity.y = 0.0f;
Bodies[i].vAngularVelocity.z = 0.0f;
Bodies[i].vAngularAcceleration.x = 0.0f;
Bodies[i].vAngularAcceleration.y = 0.0f;
Bodies[i].vAngularAcceleration.z = 0.0f;
Bodies[i].vAcceleration.x = 0.0f;
Bodies[i].vAcceleration.y = 0.0f;
Bodies[i].vAcceleration.z = 0.0f;
// Set the initial forces and moments
Bodies[i].vForces.x = 0.0f;
Bodies[i].vForces.y = 0.0f;
Bodies[i].vForces.z = 0.0f;
Bodies[i].vMoments.x = 0.0f;
Bodies[i].vMoments.y = 0.0f;
Bodies[i].vMoments.z = 0.0f;
// Zero the velocity in body space coordinates
Billiards |
385
www.it-ebooks.info
Bodies[i].vVelocityBody.x = 0.0f;
Bodies[i].vVelocityBody.y = 0.0f;
Bodies[i].vVelocityBody.z = 0.0f;
// Set the initial orientation
iRoll = 0.0f;
iPitch = 0.0f;
iYaw = 0.0f;
Bodies[i].qOrientation = MakeQFromEulerAngles(iRoll, iPitch, iYaw);
// Set the mass properties
Bodies[i].fMass = BALLWEIGHT/(-g);
Ixx = 2.0f * Bodies[i].fHass / 5.0f * (BALLDIAMETER*BALLDIAMETER);
Izz = Iyy = Ixx;
Bodies[i].mlnertia.ell = Ixx;
Bodies[i].mlnertia.el2 = 0;
Bodies[i].mlnertia.el3 = 0;
Bodies[i].mlnertia.e21 = 0;
Bodies[i].mlnertia.e22 = Iyy;
Bodies[i].mlnertia.e23 = 0;
Bodies[i].mlnertia.e31 = 0;
Bodies[i].mlnertia.e32 = 0;
Bodies[i].mlnertia.e33 = Izz;
Bodies[i].mlnertialnverse = Bodies[i].mlnertia.InverseQ;
Bodies[i].fRadius = BALLDIAMETER/2;
}
}
Stepping the Simulation
During each time step, the simulation’s main loop makes a call to StepSimulation. This
function, shown next, is almost identical to the StepSimulation functions we covered
in the other examples shown throughout this book, so there really are no surprises here.
StepSimulation first makes a call to CalcObjectForces, which we’ll discuss momen¬
tarily, and then proceeds to integrate the equations of motion for each ball. We use a
basic Euler scheme here for simplicity. After integrating, StepSimulation makes a few
function calls to deal with collisions. We’ll cover those shortly.
void StepSimulation(float dtlme)
{
Vector
Ae;
int
i;
float
dt = dtlme;
int
check = NOCOLLISION;
int
c = 0;
// Calculate all of the forces and moments on the balls:
CalcObjectForces();
386 | Chapter 19: Sports
www.it-ebooks.info
// Integrate the equations of notion:
for(i=0; i<NUMBODIES; i++)
{
// Calculate the acceleration in earth space:
Ae = Bodies[i].vForces / Bodies[i].fMass;
Bodies[i].vAcceleration = Ae;
// Calculate the velocity in earth space:
Bodies[i].vVelocity += Ae * dt;
// Calculate the position in earth space:
Bodies[i].vPosition += Bodiesfi].vVelocity * dt;
// Now handle the rotations:
float nag;
Bodies[i].vAngularAcceleration = Bodiesfi].nlnertialnverse *
(Bodies[i].vMonents -
(Bodies[i].vAngularVelocity*
(Bodies[i].nlnertia *
Bodies[i].vAngularVelocity)));
Bodies[i].vAngularVelocity += Bodiesfi].vAngularAcceleration * dt;
// Calculate the new rotation quaternion:
Bodies[i].qOrientation += (Bodies[i].qOrientation *
Bodies[i].vAngularVelocity) *
(0.5f * dt);
// Now nornalize the orientation quaternion:
nag = Bodiesfi] .qOrientation.MagnitudeQ;
if (nag != 0)
Bodies[i].qOrientation /= nag;
// Calculate the velocity in body space:
Bodies[i].vVelocityBody = QVRotate(~Bodies[i].qOrientation,
Bodies[i].vVelocity);
// Get the angular velocity in global coords:
Bodies[i].vAngularVelocityGlobal = QVRotate(Bodies[i].qOrientation,
Bodies[i].vAngularVelocity);
// Get the angular acceleration in global coords:
Bodies[i].vAngularAccelerationGlobal = QVRotate(Bodies[i].qOrientation,
Bodies[i].vAngularAcceleration);
// Get the inverse intertia tensor in global coordinates
Matrix3x3 R, RT;
R = MakeMatrixFronQuaternion(Bodies[i].qOrientation);
RT = R.Transpose();
Bodies[i].nlelnverse = R * Bodies[i].nlnertialnverse * RT;
Billiards | 387
www.it-ebooks.info
// Calculate the air speed:
Bodies[i] .fSpeed = Bodies[i]. vVelocity .MagnitudeQ;
// Get the Euler angles for our information
Vector u;
u = MakeEulerAnglesFromQ(Bodies[i].qOrientation);
Bodies[i].vEulerAngles.x = u.x; // roll
Bodies[i].vEulerAngles.y = u.y; // pitch
Bodies[i].vEulerAngles.z = u.z; // yaw
}
// Handle Collisions :
check = CheckForCollisions();
if(check == COLLISION)
ResolveCollisions();
}
Calculating Forces
The first function call made by StepSinulation is a call to CalcOb jectForces, which
is responsible for computing all the forces on each ball except collision forces. This is
the same approach used in previous examples. The entire CalcObjectForces source
code is included here:
void CalcObjectForces(void)
{
Vector Fb, Mb;
Vector vDragVector;
Vector vAngularDragVector;
tnt I, j;
Vector ContactForce;
Vector pt;
int check = NOCOLLISION;
pCollision pCollisionData;
Vector FrictionForce;
Vector fDir;
double speed;
Vector FRn, FRt;
for(i=0; i<NUMBODIES; i++)
{
// Reset forces and moments:
Bodies[i].vForces.x = 0.0f;
Bodies[i].vForces.y = 0.0f;
Bodies[i].vForces.z = 0.0f;
Bodies[i].vMoments.x = 0.0f;
Bodies[i].vMoments.y = 0.0f;
Bodies[i].vMoments.z = 0.0f;
388 | Chapter 19: Sports
www.it-ebooks.info
Fb.x = 0.0f; Mb.x = 0.0f;
Fb.y = 0.0f; Mb.y = 0.0f;
Fb.z = 0.0f; Mb.z = 0.0f;
// Do drag force:
vDragVector = -Bodiesfi].vVelocityBody;
vDragVector.Nornalize();
speed = Bodiesfi] .vVelocityBody.MagnitudeQ;
Fb += vDragVector * ((1.0f/2.0f)*speed * speed * rho *
LINEARDRAGCOEFFICIENT * pow(Bodies[i].fRadius,2) *
Bodies[i].fRadius*pi);
vAngularDragVector = -Bodies[i].vAngularVelocity;
vAngularDragVector.Normalize();
Mb += vAngularDragVector * (Bodies[i].vAngularVelocity.Magnitude() *
Bodies[i.].vAngularVelocity.Magnltude() * rho * ANGULARDRAGCOEFFICIENT
* 4 * pow(Bodies[i].fRadius,2)*pi);
// Convert forces from model space to earth space:
Bodies[i].vForces = QVRotate(Bodies[i].qOrientation, Fb);
// Apply gravity:
Bodies[i].vForces.z += GRAVITY * Bodiesfi].fMass;
// Save the moments:
Bodies[i].vMoments += Mb;
// Handle contacts with ground plane:
Bodies[i].vAcceleration = Bodiesfi].vForces / Bodies[i].fMass;
Bodies[i].vAngularAcceleration = Bodies[i].mlnertialnverse *
(Bodies[i].vMoments -
(Bodies[i],vAngularVelocity A
(Bodies[i].mlnertia *
Bodies[i].vAngularVelocity)));
// Resolve ground plane contacts:
FlushCollisionData();
pCollisionData = Collisions;
NumCollisions = 0;
if(DOCONTACT)
check = CheckGroundPlaneContacts(pCollisionData, i);
if((check == CONTACT) && DOCONTACT)
{ j = 0;
{
assert(NumCollisions <= 1);
ContactForce = (Bodies[i].fMass * (-Bodies[i].vAcceleration *
Collisions[j].vCollisionNormal)) *
Collisions[j].vCollisionNormal;
if(DOFRICTION)
Billiards | 389
www.it-ebooks.info
double vt = fabs(Colltsions[j].vRelativeVeloclty *
Collisions[j].vCollisionTangent);
if(vt > VELOCITYTOLERANCE)
{
// Kinetic:
FrictionForce = (ContactForce.MagnitudeQ *
FRICTIONCOEFFICIENTGROUND) *
Collisions[j].vCollisionTangent;
} else {
// Static:
FrictionForce = (ContactForce.MagnitudeQ *
FRICTIONCOEFFICIENTGROUND * 2 *
vt/VELOCITYTOLERANCE) *
Collisions[j].vCollisionTangent;
}
} else
FrictionForce.x = FrictionForce.y = FrictionForce.z = 0;
// Do rolling resistance:
if(Bodies[i].vAngularVelocity.MagnitudeQ > VELOCITYTOLERANCE)
{
FRn = ContactForce.MagnitudeQ *
Collisions[ j].vCollisionNomal;
Collisions [ j ]. vCollisionTangent. NornalizeQ;
Vector n = (Collisionsfj].vCollisionTangent
*(ROLLINGRESISTANCECOEFFICIENT *
Bodies[i].fRadius)) A FRn;
double nag = n.MagnitudeQ;
Vector a = Bodies[i].vAngularVelocity;
a.NornalizeQ;
Bodies[i].vMonents += -a * nag;
}
// accunlate contact and friction forces and noments
Bodies[i].vForces += ContactForce;
Bodies[i].vForces += FrictionForce;
ContactForce = QVRotate(~Bodies[l].qOrientation, ContactForce);
FrictionForce = QVRotate(~Bodies[i].qOrientation,
FrictionForce);
pt = Collisions[j].vCollisionPoint - Bodies[i].vPosition;
pt = QVRotate(~Bodies[i].qOrientation, pt);
Bodies[i].vMonents += pt A ContactForce;
Bodies[i].vMonents += pt A FrictionForce;
}
}
}
}
390 | Chapter 19: Sports
www.it-ebooks.info
As you can see, upon entering CalcObjectForces the code enters a loop that cycles
through all the billiard ball objects, computing the forces acting on each. The first force
computed is simple aerodynamic drag. Both linear and angular drag are computed. We
compute the magnitude of the linear drag by multiplying the linear drag coefficient by
l/2p V2r2nr, where p is the density of air, V is the ball’s linear speed, and r is the ball’s
radius. We compute the magnitude of the angular drag moment by multiplying the
angular drag coefficient by cop4r2Tt, where to is angular speed. Since drag retards motion,
the linear drag and angular drag vectors are simply the opposite of the linear and angular
velocity vectors, respectively. Normalizing those vectors and then multiplying by the
respective drag magnitudes yields the linear and angular drag force and moment vec¬
tors.
The next set of forces calculated in CalcObjectForces is the contact forces between the
table top and each ball. There are three contact forces. One is the vertical force that
keeps the balls from falling through the table, another is the friction force that arises as
the balls slide along the table, and the third is rolling resistance. These forces arise only
if the ball is in contact with the table. We’ll address how to determine whether a ball is
in contact with the table later in this chapter. For now, we’ll assume there’s contact and
show you how to compute the contact forces.
To compute the vertical force required to keep the ball from falling through the table,
we must first compute the ball’s linear acceleration, which is equal to the sum of forces
(excluding contact forces) acting on the ball divided by the ball’s mass. Next, we take
the negative dot product of that acceleration and the vector perpendicular to the table
surface and multiply the result by the ball’s mass. This yields the magnitude of the contact
force, and to get the vector we multiply that magnitude by the unit vector perpendicular
to the table’s surface. The following two lines of code perform these calculations:
Bodtes[t].vAcceleration = Bodles[i].vForces / Bodies[l].fMass;
ContactForce = (Bodiesfi].fMass * (-Bodiesfi].vAcceteration *
Collisions!j].vCollisionNormal)) *
Collisions!j]•vCollisionNormal;
The vCollisionNormal vector is determined by CheckGroundPlaneContacts, which
we’ll cover later. As with collisions, CheckGroundPlaneContacts fills in a data structure
containing the point of contact, relative velocity between the ball and table at the point
of contact, and the contact normal and tangent vectors, among other data.
To compute the sliding friction force, we must first determine the tangential component
of the relative velocity between the ball and table. If the ball is sliding or slipping as it
rolls, then the relative tangential velocity will be greater than 0. If the ball is rolling
without sliding, then the relative velocity will be 0. In either case, there will be a friction
force; in the former case, we’ll use the kinetic friction coefficient, and in the latter we’ll
use the static friction coefficient. Friction force is computed in the same way we showed
you in Chapter 3. The following lines of code perform all these calculations:
Billiards | 391
www.it-ebooks.info
ContactForce = (Bodiesfi].fMass * (-Bodies[1].vAcceleration *
Colllsions[j].vCollislonNormal)) *
Collisionsfj] .vCollislonNormal;
double vt = fabs(Colllslons[j].vRelatlveVeloclty *
Collisions[j].vCollisionTangent);
if(vt > VELOCITYTOLERANCE)
{
// Kinetic:
FrictionForce = (ContactForce.Magnitude() *
FRICTIONCOEFFICIENTGROUND) *
Collisions[j].vCollisionTangent;
} else {
// Static:
FrictionForce = (ContactForce.Magnitude() *
FRICTIONCOEFFICIENTGROUND * 2 *
vt/VELOCITYTOLERANCE) *
Collisions[j].vCollisionTangent;
}
Keep in mind that these forces will create moments if they do not act through the ball’s
center of gravity. So, after computing and aggregating these forces, you must also resolve
any moments created and aggregate those using the same formulas we’ve shown through
this book. The following lines of code take care of these tasks:
// accunlate contact and friction forces and moments
Bodies[i].vForces += ContactForce;
Bodies[i].vForces += FrictionForce;
ContactForce = QVRotate(~Bodies[i].qOrientation, ContactForce);
FrictionForce = QVRotate(~Bodies[i].qOrientation,
FrictionForce);
pt = Collisionsfj].vCollisionPoint - Bodies[i].vPosition;
pt = QVRotate(~Bodies[i].qOrientation, pt);
Bodiesfi].vMoments += pt A ContactForce;
Bodies[i].vMoments += pt A FrictionForce;
Rolling resistance arises by virtue of small deformations in the cloth-covered table cre¬
ating a little divot that the ball must overcome as it rolls. This divot shifts the center of
application of the contact force just a little bit in the direction of rolling. That small
offset results in a moment when multiplied by the contact force. The resulting moment
opposes rolling; otherwise, without some other resistance the ball would continue roll¬
ing unrealistically. The following code computes the rolling resistance:
// Do rolling resistance:
if(Bodies[i] .vAngularVelocity.MagnitudeQ > VELOCITYTOLERANCE)
{
FRn = ContactForce.Magnitude() *
Collisions[j].vColllsionNormal;
Collisions[j].vCollisionTangent.Normalize();
Vector m = (Collisions[j].vCollisionTangent
*(ROLLINGRESISTANCECOEFFICIENT *
392 | Chapter 19: Sports
www.it-ebooks.info
Bodles[i].fRadius)) A FRn;
double nag = m.MagnitudeQ;
Vector a = Bodles[l].vAngularVelocity;
a.NornalizeO;
Bodies[i].vMoments += -a * nag;
Handling Collisions
Earlier you saw where StepSimulation makes a few function calls to deal with collision
checking and response. You also saw where CalcObjectForces makes a function call
that checks for contacts. The functions that check for collisions or contacts make use of
the Collisions array we showed you earlier. This array stores all the relevant informa¬
tion pertaining to collisions or contacts—the collision or contact manifold, normal and
tangent vectors, relative velocity, etc.
The first function we’ll consider is CheckForCollisions, which is called toward the end
of StepSimulation. CheckForCollisions checks for ball-ball collisions; we have a sep¬
arate function to check for ball-table collisions that we’ll get to later. CheckForColli
sions relies on concepts we’ve already discussed and showed you in earlier chapters, so
we’ll summarize its action here. Basically, two billiard balls are colliding if 1) they are
headed toward each other, and 2) the distance separating their centers is less than or
equal to the sum of their radii. Ifboth of these criteria are met, then a collision is recorded
and all relevant data is stored in the Collisions array:
int
1
CheckForCollisions(void)
int
int
Vector
pCollision
int
float
float
Vector
status = NOCOLLISION;
i, j;
d;
pCollisionData;
check = NOCOLLISION;
r;
s;
trip;
FlushCollisionData();
pCollisionData = Collisions;
NupiCollisions = 0;
// check object collisions with each other
for(i=0; i<NUMBODIES; i++)
{
for(j=0; j<NUMBODIES; j++)
if((j!=i) && !CollisionRecordedAlready(i, j))
{
// do a bounding sphere check
d = Bodiesfi].vPosition - Bodies!j].vPosition;
r = Bodies[i].fRadius + Bodies[j].fRadius;
s = d.Magnitude!) - r;
Billiards | 393
www.it-ebooks.info
if(s < COLLISIONTOLERANCE)
{// possible collision
Vector ptl, pt2, veil, vel2, n, Vr;
float Vrn;
ptl = (Bodies[i].vPosition + Bodies[j].vPosition)/2;
tmp = pt2 = ptl;
ptl = ptl-Bodies[i].vPosition;
pt2 = pt2-Bodies[j].vPosition;
veil = Bodiesfi].vVelocity +
(Bodies[i].vAngularVelocityGlobal A ptl);
vel2 = Bodies[j].vVelocity +
(Bodies[j].vAngularVelocityGlobal A pt2);
n = d;
n.Normalize();
Vr = (veil - vel2);
Vrn = Vr * n;
if(Vrn < -VELOCITYTOLERANCE)
{
// Have a collision so fill the data structure
assert(NumCollisions < (NUMB0DIES*8));
if(NumCollisions < (NUMB0DIES*8))
{
pCollisionData->bodyl = i;
pCollisionData->body2 = j;
pCollisionData->vCollisionNormal = n;
pCollisionData->vCollisionPoint = tmp;
pCollisionData->vRelativeVelocity = Vr;
pCollisionData->vCollisionTangent = (n A Vr) A n;
pCollisionData->vCollisionTangent.Normalize();
pCollisionData++;
NumCollisions++;
status = COLLISION;
for(i=0; i<NUMBODIES; i++)
{
check = NOCOLLISION;
assert(NumCollisions < (NUMB0DIES*8));
check = CheckGroundPlaneCollisions(pCollisionData, i);
if(check == COLLISION)
{
394 | Chapter 19: Sports
www.it-ebooks.info
status = COLLISION;
pColltsionData++;
NumCollisions++;
}
}
return status;
}
Since CheckForCollisions loops through all of the balls checking for collisions with
every other ball, it is possible that a collision would be recorded twice. For example, the
ith ball may be found to be colliding with the jth ball, and later the jth ball would also
be found to be colliding with the ith ball. We don’t want to record that information twice,
so we use the following function to check if a collision between two particular balls is
already recorded. If so, we skip re-recording the data:
bool CollisionRecordedAlready(int i, int j)
1
int k;
int bl, b2;
for(k=0; k<NumCollisions; k++)
{
bl = Collisionsfk].bodyl;
b2 = Collisionsfk].body2;
if( ((bl == i) && (b2 == j)) | |
((bl == j) && (b2 == i» )
return true;
}
return false;
1
Checking ball-table collisions is fairly straightforward as well. If 1) a ball is found to be
headed toward the table with some velocity greater than 0 (or some small threshold),
and 2) the ball’s vertical position to its center is less than or equal to its radius, then we
record a collision. CheckGroundPlaneCollisions handles this for us:
int CheckGroundPlaneCollisions(pCollision CollisionData, int bodyl)
1
Vector
top;
Vector
veil;
Vector
ptl;
Vector
Vr;
float
Vrn;
Vector
n;
int
status = NOCOLLISION;
if(Bodiesfbodyl].vPosition.z <= (Bodiesfbodyl].fRadius))
1
ptl = Bodiesfbodyl].vPosition;
Billiards | 395
www.it-ebooks.info
ptl.z = COLLISIONTOLERANCE;
tnp = ptl;
ptl = ptl-Bodies[bodyl].vPosition;
veil = Bodies[bodyl].vVeloclty/*Body*/ +
(Bodiesfbodyl].vAngularVelocityGlobal A ptl);
n.x = 0;
n.y = 0;
n.z = 1;
Vr = veil;
Vrn = Vr * n;
if(Vrn < -VELOCITYTOLERANCE)
{
// Have a collision so fill the data structure
assert(NutnCollisions < (NUMBODIES*8));
if(NunCollisions < (NUMB0DIES*8))
{
CollisionData->bodyl = bodyl;
CollisionData->body2 = -1;
CollisionData->vCollisionNormal = n;
CollisionData->vCollisionPoint = trip;
CollisionData->vRelativeVelocity = Vr;
CollisionData->vCollisionTangent = (n A Vr) A n;
CollisionData->vCollisionTangent.Reverse();
CollisionData->vCollisionTangent.Normalize();
status = COLLISION;
}
}
}
return status;
}
Resolving collisions, whether ball-ball or ball-table collisions, uses the same approach
we’ve already shown you. Thus, we won’t go over the code again, and will instead just
show you the function that implements collision response:
void ResolveCollisions(void)
{
int i;
double j;
Vector ptl, pt2, vBlV, vB2V, vBlAV, vB2AV;
float fCr = COEFFICIENTOFRESTITUTION;
int bl, b2;
float Vrt;
float PluB = FRICTIONCOEFFICIENTBALLS;
float nuG = FRICTIONCOEFFICIENTGROUND;
bool dofriction = DOFRICTION;
396 | Chapter 19: Sports
www.it-ebooks.info
for(i=0; i<NumCollisions; i++)
{
bl = Collisions[i.] .bodyl;
b2 = Colllsions[l],body2;
If( (bl != -1) && (bl != b2) )
{
if(b2 != -1) // not ground plane
{
ptl = Collisions[l].vColllsionPoint - Bodies[bl].vPositlon;
pt2 = Collisions[l].vColllsionPoint - Bodies[b2].vPosition;
// Calculate impulse:
j = (-(1+fCr) * (Collisions[i].vRelativeVelocity *
Collisions[i].vCollisionNormal)) /
((l/Bodies[bl].fMass + l/Bodies[b2].fMass) +
(Collisions[i].vCollisionNormal * ( ( (ptl A
Collisions[i].vCollisionNormal) *
Bodies[bl].mlelnverse ) A ptl) ) +
(Collisions[i].vCollisionNormal * ( ( (pt2 A
Collisions[i].vCollisionNormal) *
Bodies[b2].mlelnverse ) A pt2) ) );
Vrt = Collisions[i].vRelativeVelocity *
Collisions[i].vCollisionTangent;
if(fabs(Vrt) > 0.0 && dofriction) {
Bodies[bl].vVelocity +=
((j * Collisionsfi].vCollisionNormal) +
((muB * j) * Collisions[i].vCollisionTangent)) /
Bodies[bl].fMass;
Bodies[bl].vAngularVelocityGlobal +=
(ptl A ((j * Collisions[i].vCollisionNormal) +
((muB * j) * Collisions[i].vCollisionTangent)))
Bodies[bl].mlelnverse;
Bodies[bl].vAngularVelocity =
QVRotate(-Bodies[bl].qOrientation,
Bodies[bl].vAngularVelocityGlobal);
Bodies[b2].vVelocity +=
((-j * Collisions[i].vCollisionNormal) + ((muB *
j) * Collisions[i].vCollisionTangent)) /
Bodies[b2].fMass;
Bodies[b2].vAngularVelocityGlobal +=
(pt2 A ((-j * Collisions[i].vCollisionNormal) +
((muB * j) * Collisions[i].vCollisionTangent)))
* Bodies[b2].mlelnverse;
Bodies[b2].vAngularVelocity =
QVRotate(~Bodies[b2].qOrientation,
Bodies[b2].vAngularVelocityGlobal);
Billiards |
397
www.it-ebooks.info
} else {
// Apply impulse:
Bodies[bl].vVelocity +=
(j * Collisions[i].vColllsionNomal) /
Bodies[bl].fMass;
Bodies[bl].vAngularVelocityGlobal +=
(ptl A (j * Collisions[i].vCollisionNormal)) *
Bodies[bl].mlelnverse;
Bodies[bl].vAngularVelocity =
QVRotate(-Bodies[bl].qOrientation,
Bodies[bl].vAngularVelocityGlobal);
Bodies[b2].vVelocity -=
(j * Collisions[i] .vColllsionNomal) /
Bodies[b2].fMass;
Bodies[b2].vAngularVelocityGlobal -=
(pt2 A (j * Collisions[i] .vColllsionNomal)) *
Bodies[b2].mlelnverse;
Bodies[b2].vAngularVelocity =
QVRotate(-Bodies[b2].qOrientation,
Bodies[b2].vAngularVelocityGlobal);
} else { // Ground plane:
fCr = COEFFICIENTOFRESTITUTIONGROUND;
ptl = Collisions[i].vCollisionPoint - Bodies[bl].vPosition;
// Calculate impulse:
j = (-(1+fCr) * (Collisions[i].vRelativeVelocity *
Collisionsfi].vCollisionNormal)) /
( (l/Bodies[bl].fMass) +
(Collisions[i].vCollisionNormal *
( ( (ptl A Collisions[i].vCollisionNormal) *
Bodies[bl].mlelnverse ) A ptl)));
Vrt = Collisions[i].vRelativeVelocity *
Collisions[i].vCollisionTangent;
if(fabs(Vrt) > 0.0 && dofriction) {
Bodiesfbl].vVelocity +=
( (j * Collisions[i].vCollisionNormal) + ((muG *
j) * Collisions[i].vCollisionTangent) ) /
Bodies[bl].fMass;
Bodies[bl].vAngularVelocityGlobal +=
(ptl A ((j * Collisions[i].vCollisionNormal) +
((muG * j) * Collisions[i].vCollisionTangent))) *
Bodies[bl].mlelnverse;
Bodies[bl].vAngularVelocity =
QVRotate(-Bodies[bl].qOrientation,
Bodies[bl].vAngularVelocityGlobal);
} else {
398 | Chapter 19: Sports
www.it-ebooks.info
// Apply impulse:
Bodies[bl].vVelocity +=
(j * Collislons[i].vCollisionNomal) /
Bodtes[bl].fMass;
Bodies[bl].vAngularVelocityGlobal +=
(ptl A (j * Collisions[i] .vCollisionNomal)) *
Bodies[bl].mlelnverse;
Bodies[bl].vAngularVelocity =
QVRotate(~Bodies[bl].qOrientation,
Bodies[bl].vAngularVelocityGlobal);
The final function we need to showyou is CheckGroundPlaneContacts. Recall that this
function is called from CalcObjectForces in order to determine if a ball is in resting
contact with the table. If the ball’s vertical position is less than or equal to its radius plus
some small tolerance, and if the ball’s vertical velocity is 0 (or nearly so within some
small tolerance), then we consider the ball in contact with the table. If there’s a contact,
the relevant data gets stored in the Collisions array and used to resolve the contact,
not the collision, in CalcObjectForces:
int CheckGroundPlaneContacts(pCollision CollislonData, int bodyl)
{
Vector
vl[8];
Vector
tmp;
Vector
u, v;
Vector
f[4];
Vector
veil;
Vector
ptl;
Vector
Vr;
float
Vrn;
Vector
n;
int
status = NOCOLLISION
Vector
Ar;
float
Arn;
lf(Bodies[bodyl].vPosltion.z <= (Bodiesfbodyl].fRadius + COLLISIONTOLERANCE))
{
ptl = Bodlesfbodyl].vPosition;
ptl.z = COLLISIONTOLERANCE;
tmp = ptl;
ptl = ptl-Bodles[bodyl].vPosition;
veil = Bodles[bodyl].vVeloclty/*Body*/ +
(Bodlesfbodyl].vAngularVelocltyGlobal A ptl);
n.x = 0;
n.y = 0;
Billiards | 399
www.it-ebooks.info
n.z = 1;
Vr = veil;
Vrn = Vr * n;
if(fabs(Vrn) <= VELOCITYTOLERANCE) // at rest
{
// Check the relative acceleration:
Ar = Bodies[bodyl].vAcceleration +
(Bodies[bodyl].vAngularVelocityGlobal A
(Bodies[bodyl].vAngularVelocityGlobal A ptl)) +
(Bodies[bodyl].vAngularAccelerationGlobal A ptl);
Arn = Ar * n;
if(Arn <= 0.0f)
{
// We have a contact so fill the data structure
assert(NunCollisions < (NUMB0DIES*8));
if(NumCollisions < (NUMB0DIES*8))
{
CollisionData->bodyl = bodyl;
CollisionData->body2 = -1;
CollisionData->vCollisionNornal = n;
CollisionData->vCollisionPoint = trip;
CollisionData->vRelativeVelocity = Vr;
CollisionData->vRelativeAcceleration = Ar;
CollisionData->vCollisionTangent = (n A Vr) A n;
CollisionData->vCollisionTangent.Reverse();
CollisionData->vCollisionTangent .NortnalizeO;
CollisionData++;
NumCollistons++;
status = CONTACT;
}
}
}
}
return status;
}
That’s all there is to this billiards example. As you can see, we used substantially the
same methods shown in other examples throughout this book to implement this ex¬
ample. About the only new information we’ve shown here is how to compute rolling
resistance. With a little effort, you can combine the material presented in this example
with the projectile motion material presented in Chapter 6 to model all sorts of sports
balls. Whether you’re modeling a billiard ball bouncing off a table or a basketball
bouncing off a backboard, the methods are the same. The only things that will change
are the empirical coefficients you use to model each ball and surface. Have fun.
400 | Chapter 19: Sports
www.it-ebooks.info
PART IV
Digital Physics
Part IV covers digital physics in a broad sense. This is an exciting topic, as it relates to
the technologies associated with mobile platforms, such as smartphones like the iPhone,
and groundbreaking game systems like the Wii. Chapters in this part of the book will
explain the physics behind accelerometers, touch screens, GPS, and other gizmos,
showing you how to leverage these elements in your games. We recognize that these
topics are not what most game programmers typically think about when they think of
game physics; however, these technologies play an increasingly important role in
modern mobile games, and we feel it’s important to explain their underlying physics in
the hope that you’ll be better able to leverage these technologies in your games.
www.it-ebooks.info
www.it-ebooks.info
CHAPTER 20
Touch Screens
It is hard to deny that we are currently moving toward a post-PC computing environ¬
ment. The proliferation of smartphones, tablets, and other mobile computing platforms
will have far-reaching implications for how people interact with computers. These form
factors do not allow for the more traditional mouse and keyboard of input for games
and therefore rely heavily on the use of touch screens. This chapter aims to give you
some background on the different types of touch screens, how they work, and their
technical limitations. Note that we will extend our particle simulator to work with the
iPhones capacitive touch screen; the final product is very similar to the mouse-driven
version but provides a starting point for a touch-driven physics simulator.
While this chapter will primarily deal with the most two most common types of touch-
sensitive screens, resistive and capacitive, the following section gives an overview of
many different types. In the future we may see a move to more exotic devices, especially
for large-format computing devices.
Types of Touch Screens
Resistive
Resistive touch screens are basically a giant network of tiny buttons. Some of them have
4,096 x 4,096 buttons in a single square inch! OK, so they are not quite just normal
buttons, but they come close. Resistive touch screens have at least two layers of con¬
ductors with an air gap between them. As you press on the screen, you close the gap.
Bam, circuit complete, button pressed. We will flesh out that simplified description
shortly.
403
www.it-ebooks.info
Capacitive
Also a topic we’ll soon discuss in detail, capacitive touch screens are very common on
todays smartphones. These touch screens operate by calculating the change in electrical
capacitance at the four corners of the screen when your finger influences the capacitive
nature of the circuits under the glass. The limitation is that whatever touches the screen
must be electrically conductive. If you insulate your fingers with gloves, the screen will
no longer be able to locate your touch. However, this can be solved with a few stitches
of conductive thread.
Infrared and Optical Imaging
Infrared touch screens use arrays of infrared LED and photodetectors to detect and
interpret an object breaking the path of a LED-photodetector pair. This uses line¬
scanning techniques and is a very robust design.
Optical imaging techniques are relative newcomers to the touch screen scene whose big
advantage is that they are extremely scalable. They use imaging devices and light sources
to detect where the screen is being touched by interpreting any shadows cast by an object
through the thickness of the material.
Exotic: Dispersive Signal and Surface Acoustic Wave
Several other exotic touch screen technologies exist. We won’t get into detail here, but
3M has a system that detects mechanical energy in glass caused by a touch. The amount
of vibration energy that reaches each sensor determines the position.
Another example of exotic screen input, surface acoustic wave technology detects
changes in the pattern of ultrasonic waves traveling along the surface of the screen.
Step-by-Step Physics
Resistive Touch Screens
Resistive touch screens are classified as a passive touch screen technology because the
screen registers a touch without any active participation by the object touching the
screen. This is their major benefit over active technologies, such as capacitive touch
screens, as resistive screens can be activated by nonconductive objects like a pen or
gloved finger. In the past, resistive screens were limited to a single input, and that’s the
type we’ll describe, but they can be made to work with two or more simultaneous inputs,
also known as multitouch.
404 | Chapter 20:Touch Screens
www.it-ebooks.info
One-dimensional resistive touch sensor
To ease ourselves into this discussion, we will begin by looking at a one-dimensional
touch screen. Let’s imagine we have built the machine described in Figure 20-1.
Figure 20-1. Linear resistive touch sensor
As you can see, our sensor has two states, an open circuit state and a closed circuit
state. In the open circuit state, the controller is supplying a 5 V signal to pin 2 and waiting
for any return voltage on pin 1. With no touch to bring the wires together, the circuit
is open. No voltage is present at pin 1, and therefore no touch is sensed. When the wires
are touched, they are brought together and the circuit is closed. A voltage will then be
present at pin 1. A touch event is registered.
This type of sensor, which looks only for the presence or absence of voltage without
regard to its value, is called a digital sensor. It can detect only two states: on or off (1 or
0, respectively). OK, so it’s not quite a touch screen yet; essentially at this point all we
have is a simple button. Moving forward, let’s say that we not only want to trigger an
event when we press our button, but we also want to simultaneously input a value based
on the location along the wire, L, that we pressed.
To accomplish this, the controller patiently waits for a voltage at pin 1. When it senses
a voltage, that digital “on” switch causes the controller to then probe the voltage that is
present, which we have labeled V x . Now we get to the reason it is called a resistive touch
sensor. Current, voltage, and resistance are all interrelated by Ohm’s law. This physical
relationship is expressed as:
V = IR
Where V is voltage, I is current, and R is resistance. The exact physical meaning of each
of these is less important right now than how they are related, so don’t get too worried
if you can’t recall their definitions. In the case of our circuit, I, or current, is going to be
Step-by-Step Physics | 405
www.it-ebooks.info
constant. As the controller measures the voltage, V, at pin 1, we can now solve for
resistance:
R = V/l
With our constant current, I, and our known voltage, V x , we can calculate the resistance
of the circuit by inputting (5 - V x ) for V as follows:
R = (5-Vx)/I(l)
Note that we have to have the change in voltage across our resistor (the wire), so be sure
to use the difference between the two values. Every conductor has an inherent internal
resistance, and through testing we can determine what the resistance is, measured in
ohms per unit length, and use that to determine our total resistance described by:
R = 2rL
where r is the aforementioned ohms per unit length and R is the total resistance of our
circuit. Note that we have multiplied L by 2 to account for both the wire run to the point
of contact and back. If we substitute this for R in our earlier equation, we now have:
2rL = (5 - Vx) /1
And finally:
L = (5 - Vx) / (2rl)
where L is the only unknown. To illustrate, let us assume the measured voltage is 4.95V
and the wires are 24-gauge copper wires. A quick look in a standard electrical engi¬
neering book gives the resistance as 0.08422 ohms per meter. When we designed our
constant current source, let’s say we picked 50 milliamps:
L = (5 - 4.95V) / ((2)(.08422 ohms/meter)(.05A))
L = 5.9 meters from the controller
As you can see, the material’s resistance per meter, the constant current supplied, and
the sensitivity of the voltage-sensing circuit must all be finely tuned to ensure that the
controller is capable of sensing touch events in the appropriate dimensions. In a resistive
touch screen, the wires are microscopic so that the resistance per meter is much higher.
This allows the screen to detect smaller distances.
Four-wire resistive touch screen
With some modification, we can expand our previous model to two dimensions. In the
four-wire touch screen, there are four basic layers and four wires, three of which will be
406 | Chapter 20:Touch Screens
www.it-ebooks.info
used at any given time. The general layout is given in Figure 20-2. The squares containing
the X and Y wires would actually be overlapped but are shown skewed here for clarity.
Figure 20-2. Four-wire touch screen
The reason for calling it a. four-wire touch screen should now be obvious; however,
remember that only three of the wires will be active at any time. The basic structure is
shown in Figure 20-3.
FLEXIBLE CONDUCTOR -Y AXIS_
_ FLEXIBLE INSULATING GAP
SOLID CONDUCTOR-X AXIS _+5v
GLASS
Figure 20-3. Four-wire touch screen profile
The first layer of the screen comprises a flexible conductor separated by an insulating
gap. Under the gap lies a solid conductor. When a finger presses down on the outer layer
of flexible conductor, it crosses the gap and makes contact with the solid conductor. The
conductors are thin layers of indium-tin oxide (ITO) with silver bus bars on either end
of the sheet, shown as black lines in Figure 20-2.
Step-by-Step Physics | 407
www.it-ebooks.info
To condense the description of its operation, we’ve outlined the three possible states in
Table 20-1.
Table 20-1. Possible states for four-wire touch screen
1 Activity
Pin 1
Pin 2
Pin 3
Pin 4 1
Waiting for touch detection
Open
Open
Digital input [pullup]
Ground
Read A position
Voltage probe
Ground
Voltage source
Open
Read K position
Voltage source
Open
Voltage probe
Ground
Voltage probe means the chip is sensing the voltage on that pin, voltage source is the pin
supplying a voltage to ground, and open means it is unused. The sequence of a touch
event begins with pin 1 and pin 2 open. Pin 3 is configured to digital input with pullup
signifying a voltage is supplied to the pin. When a finger presses on the outer layer and
makes contact with the lower layer, pin 3 goes to ground. When the controller senses
the voltage fall on pin 3, it moves to the second row and reads the X position.
To read the X position, the lower layer is energized from pin 3 to pin 2. The voltage
source creates a gradient along the layer. Pin 1, connected to the upper layer, delivers a
voltage to the controller when a touch pushes it down to make contact with the lower
energized layer. The value of this voltage depends on where the contact is made in the
gradient, much like the previous linear example. Once the X position is known, the
controller moves to the next row and reads the Y position.
The method of obtaining the Y position is much the same but in reverse. The voltage
supply is switched to pin 1, which develops a voltage gradient with pin 4. Then pin 3 is
probed and the voltage corresponds to the distance along the voltage gradient. As the
controller is capable of repeating the detect, read X, and read Y cycles approximately
500 times a second, the user is not aware that the screen doesn’t actually register the X
and Y coordinates at the same time.
While the four-wire resistive touch screen is the simplest two-dimensional touch sensor,
there are issues with durability. The main drawback of this type of touch screen is that,
because the layers must be separated by an insulating gap, at least one of the layers must
be flexible. In the four-wire type, the constant flexing of the first conductive layer in¬
troduces microcracks in the coating, which lead to nonlinearities and reduce the accu¬
racy. Other models of resistive touch screens overcome this issue with additional layers
that remove the need for the flexible conductor. They have also been adapted to provide
multitouch capability. We will discuss multitouch and how it works with capacitive
touch screens in greater detail shortly.
Capacitive Touch Screens
A capacitive screen uses a piece of glass coated in a transparent conductor. When your
finger or other conductor comes into contact with the screen, the electrostatic field is
408 | Chapter 20: Touch Screens
www.it-ebooks.info
disturbed, causing a change in the capacitance. To understand how capacitive screens
work, let’s quickly review capacitance in general.
A capacitor in its simplest form is two conductors, usually thin plates, separated by an
insulator. If you apply a voltage across the two conductors, a current will flow and charge
will build up. Once the voltage across the plates is equal to the supply voltage, the current
will stop. The amount of charge built up in the plates is what we measure as the capac¬
itance. Previously, we noted that one issue with resistive screens is that one part must
always flex to close a insulating gap and complete a circuit. This repetitive action even¬
tually leads to mechanical failure. A capacitor can be dynamically formed by any two
conductors separated by an insulator. Noting that glass is a good insulator, it is easy to
see that a finger separated from a conductor by glass can change the capacitance of a
system. In this way, the finger or stylus doesn’t have to cause any mechanical action, yet
it can still effect changes to the sensors, which are then used to measure the location of
the touch.
The methods of determining location based on capacitance on mobile devices are self¬
capacitance and mutual capacitance.
Self-capacitance
Anyone who has lived in a dry winter has felt the shock of a static electricity discharge.
This zap is possible because the human body is a pretty good capacitor with a capacitance
of about 22 pico-farads. This property is known as body capacitance. Self-capacitance
screens take advantage of a physical property defined by the amount of electrical charge
that must be added to an isolated conductor to raise its potential by one volt. When the
fingers act as a conductor of the body’s inherent capacitance, the sensors on the other
side of the glass experience a rise in electric potential. Given that the sensors are on the
other side of a good insulator, glass, there won’t actually be any discharge of energy,
unlike when you touch your metal car door and get “zapped.” Self-capacitance in this
manner produces a very strong signal but lacks the ability to accurately resolve multiple
touches. Therefore, it is often used in conjunction with the next type of touch screen
we’ll discuss, mutual capacitance.
Mutual capacitance
The other form of capacitance-sensing screen, mutual capacitance, is formed by a grid
of independent capacitors. A probing charge is sent over the rows or columns. As the
capacitors charge and discharge, the system can sense the capacitance of each individual
capacitor. As just discussed, the body is a good capacitor, and bringing part of it close
to the capacitor grid changes the local electric field. Those capacitors that are under a
finger or other conductor will read lower values than normal. Each capacitor can be
scanned independently, enabling high resolution of where the touch event is occurring.
Additionally, because they act independently of one another, it is possible to accurately
register multiple touches. Think of this system as taking a picture of the capacitance on
Step-by-Step Physics | 409
www.it-ebooks.info
the skin of the screen. Using algorithms similar to image processing and edge detection,
this system can compute the extent of a touch event.
Example Program
Included in the source code accompanying this book is an example of the particle ex¬
plosion program from Chapter 8 that uses touch screen input instead of a mouse click.
The code for a Cocoa touch Objective-C event is as follows:
- (void)touchesBegan:(NSSet *)touches withEvent:(UIEvent *)event
{
UITouch* touch = [[event touchesForViewrself] anyObject];
firstTouch = [touch tocationlnView:self];
self.status = YES;
[self trigger];
}
where firstTouch is defined by CGPoint firstTouch; in the header file. The CGPoint
is a Cocoa touch object capable of storing an (x,y) coordinate in the display view’s co¬
ordinate system. We can then use firstTouch. x and firstTouch. y later in our program
to give a location to the particle explosion.
As you can see, it is very similar to a mouse-based event. One big difference is that you
could adapt the code to handle multitouch events. Computers recognize only one mouse
cursor at a time, but with a touch screen you can register multiple clicks simultaneously.
Multitouch
In iOS you must first enable delivery of multiple touch events by setting the multiple
TouchEnabled property of your view to YES; the default is NO. Next you must create a
class to keep track of multiple TParticleExplosion structures. Then it is as simple as
polling the position of the start points of multiple touches to trigger multiple explosions.
The Objective C code to store the start points of multiple touch events would look like:
- (void)storeTouchPoints:(NSSet *)touches{
If ([touches count] > 0) {
for (UITouch *touch in touches) {
CGPoint *point = (CGPoint
*)CFDtctionaryGetValue(touchBeginPoints, touch);
if (point == NULL) {
point = (CGPoint *)malloc(sizeof(CGPoint));
CFDictionarySetValue(touchBeginPoints, touch, point);
}
*point = [touch tocationlnView:view.superview];
}
}
410 | Chapter 20:Touch Screens
www.it-ebooks.info
where CFDictionaryRef is an immutable dictionary object that allows the copying of
the object and its key value. One last consideration for this example is that as you are
now creating multiple physics simulations simultaneously, you may have to decrease
the frequency of the time steps to allow the animations to proceed smoothly. Multitouch
can become programmatically complex, but the physics are pretty simple. The event¬
handling guide for your particular development language should give detailed guidance
on handling the event chain.
Other Considerations
One of the major advantages of touch screens is that their layout and actions are entirely
software based. That is to say, if a certain button does not pertain to the current layout,
it can be discarded and the freed space used for additional relevant controls. We will
discuss other less obvious considerations in the following sections.
Haptic Feedback
The flip side of the advantage of not being locked into a set of physical buttons is that
the user must rely almost entirely on vision to interact with the controls. At least one of
the authors of this book uses a keyboard with no letters on it at all, instead relying totally
on the physical position of the keys to determine which key to strike. This would be
much harder without the tactile and audio cues to signify that the correct key has been
pressed. Indeed, it is easy to tell when he is typing poorly because the backspace key is
much louder than the rest!
The method of including physical feedback to assist a user in interaction with entirely
virtual objects is known as haptic feedback. The first use of haptic feedback in games
was limited to arcade games such as Motocross, in which the handlebars shook after an
in-game impact. It is now considered standard on video game controllers, which vibrate
to inform the user of some event.
In the realm of touch screens, haptic feedback is used to inform the user of a successful
key strike or other touch-based event. Some touch screens even incorporate some
movement of the entire screen when pressed. This feedback still doesn’t allow touch
typing, as it only dynamically responds and provides no static tactile feedback for dif¬
ferent buttons.
Modeling Touch Screens in Games
Given their planar nature and the lack of inherent haptic feedback, touch screens can
be an easy way to implement controls with which a character in a game can interact.
The amount of physical modeling required to create a realistic in-game keyboard is
pretty intense. Thus, there are very few examples of an in-game character having to sit
down and type a code in to a terminal via a standard keyboard.
Other Considerations | 411
www.it-ebooks.info
By using touch screens for in-play control of objects, you can avoid an additional phys¬
ical model while retaining realism. It would also be interesting to see games use realistic
touch screen interfaces so that a character would have to remove his gloves to use a
capacitive screen. Lastly, the exotic screen technologies mentioned earlier provide many
creative avenues of modeling those types of screens in games. For example, for screens
measuring sound waves in the glass or other mechanical energy, low-grade explosions
could be used to trigger these in-game input devices.
Difference from Mouse-Based Input
One important consideration for game developers in regards to touch screens is the
difference from traditional mouse- and keyboard-based gaming. As console game de¬
velopers have long been aware, it is hard to compete with the speed and accuracy of the
mouse/keyboard combination. Many first-person shooters segregate their online gam¬
ing between controllers and mouse/keyboard setups, as the accuracy and speed of the
mouse gives those players an unfair advantage. Upon using touch screens on many
different gaming devices and mobile computing platforms, we feel that this advantage
is even more pronounced.
A touch by a finger is an elliptical shape whose contact patch depends on the specific
finger being used, the pressure applied, and the orientation of the finger. The user gen¬
erally perceives the point of touch to be below where the actual center of contact is, so
adjustments must be made. This is generally all handled automatically by the operating
system so that a single touch point is computed and handed to the game via an API.
However, this generic approach to computing touches must obviously sacrifice accuracy
for universality so that it is not calibrated for one specific user.
Another inherent drawback to touch screens is the need to touch the screen. This means
a large portion of your hand will be blocking the screen when you are controlling that
element. One can imagine that in a first-person shooter, this would be a great disad¬
vantage over someone who is playing with a keyboard and mouse.
Lastly, mouseover is not available to touch-screen-based input. Consider a game where
you would trigger actions by merely moving a mouse cursor over an object. These
actions could be distinct from clicking on the same object. However, with touch-screen-
based input, that object would be obscured by whatever is triggering the screen, there¬
fore rendering the mouseover action invisible to the user.
Custom Gestures
As a last note, another possibility for touch input to a game is the use of custom gestures.
These allow the user to draw a shape on the screen that the program recognizes as a
gesture. It can then execute arbitrary code based on that input. As this is more pattern
recognition then physics, we won’t cover it here, but we can recommend the book
Designing Gestural Interfaces by Dan Saffer (O’Reilly) as a detailed look at this subject.
412 | Chapter 20: Touch Screens
www.it-ebooks.info
CHAPTER 21
Accelerometers
Accelerometers are a good introduction to a class of electronic components called
microelectromechanical systems (MEMS). An accelerometer can either be one-axis, two-
axis, or three-axis. This designates how many different directions it can simultaneously
measure acceleration. Most gaming devices have three-axis accelerometers.
As far as game development is concerned, acceleration values are typically delivered to
your program via an API with units in multiples of g. One g is equal to the acceleration
caused by gravity on the Earth, or 9.8 m/s 2 . Let’s pretend that we have a one-axis accel¬
erometer and we orient it such that the axis is pointing toward the center of the earth.
It would register \g. Now, if we travel far away from any mass, such that there is no
gravity, the accelerometer will read 0. If we then accelerate it such that in one second it
goes from 0 m/s to 9.8 m/s, the accelerometer will read a steady Ig during that one-
second interval. Indeed, it is impossible to tell the difference between acceleration due
to gravity and acceleration due to changing velocity.
Real-life motion is generally nonsteady. Depending on your application’s goals, you
might have to apply different smoothing functions such as high-pass or low-pass fil¬
ters. This amounts to digital signal processing, a topic that has consumed entire texts.
One example we can recommend is Digital Signal Processing: A Computer Science Per¬
spective by Jonathan Y. Stein (Wiley).
Also, many accelerometers have a method to set the polling rate, or the number of times
per second that the program requests updates from the accelerometer. This is called
frequency and is given in hertz (Hz). This parameter can be used to enhance the per¬
formance of the program when fine resolution of the acceleration over time is not
needed.
When you accept input from an accelerometer—or do any other kind of signal pro¬
cessing—you have to accept that input won’t come precisely when you want it. The
operating systems normally used for gaming—Windows, OS X, Linux—are not real-
413
www.it-ebooks.info
time environments. This means that although you set the polling rate at once a second,
this guarantees only that the data will be delivered no sooner than once a second. If
something distracts the operating system, such as the arrival of packets on the network,
the signal you get from the accelerometer may be delayed.
Accelerometer Theory
The way in which MEMS measure accelerometers is more basic in principle than you
may think. The major accomplishment is miniaturizing the technology until it can fit
inside a computer chip! To clearly illustrate the basic principle, we will first show you
the mechanics of it in the macro-scale version of a known mass and spring. Let’s say
you build something like the contraption shown in Figure 21-1 and take it on an elevator
in an area where there is no gravity. We’ll worry about the effects of gravity in a minute.
Figure 21-1. Simple accelerometer in absence of acceleration
414 | Chapter 21: Accelerometers
www.it-ebooks.info
As you can see, the device consists of a known mass at the end of a spring next to a
measuring stick. When the elevator is not accelerating, the mass is at the 0 mark. When
the elevator accelerates up or down, the mass at the end of the spring resists that accel¬
eration and tends to stay at rest. This is Newton’s first law in action. Inertial loading
causes the spring to stretch or compress. If the elevator is accelerating upward, the mass
will cause the spring to stretch downward. Recall from Chapter 3 that the force acting
on a spring is linearly dependent on the displacement of the mass via the equation:
F n = kd
We can directly measure the displacement, d, so we can determine the force in that
direction, n. As the mass is known, voila!
a n = m/F n
As an aside, the fact that you can tell that you are accelerating without having to look
outside the elevator is what makes this a “noninertial frame of reference,” as discussed
in Chapter 2. No worries if you don’t totally understand that; it isn’t important for what
we are discussing here.
Now, let’s put our elevator back on earth. With the same device, the mass will not be at
0 even if the elevator is not accelerating because gravity is pulling it down. Previously
we used units of inches, which we then converted to force and finally to acceleration.
However, we now have a direct measure of the acceleration due to gravity and could
easily place a mark on where the mass is and call it Ig. Also we could place marks along
the ruler at the same intervals. Now, we accelerate the elevator upward at 9.8 m/s 2 . The
mass should move down the scale to 2 g, and anyone standing in the elevator would feel
twice as heavy as normal.
Let’s say we wanted to accelerate the elevator downward at 9.8 m/s 2 . We could easily do
this by just releasing the brakes and letting gravity do the work. Now in freefall we don’t
feel gravity at all, right? That’s because the downward acceleration is canceling the ac¬
celeration due to gravity. The mass will be back at 0 just like out in space, far from any
gravitational bodies. It is for this reason, and not a lack of gravity, that astronauts float
around. They are in freefall around the earth.
Lastly, if we accelerate the elevator downward at 2 g, the mass would move to the -1 g
mark on the ruler. This is because the downward acceleration is now overwhelming
gravity. Those in the elevator would find themselves standing on the ceiling feeling
exactly as they would standing on the ground! In fact, one of Einstein’s accomplishments
was showing that it is impossible to distinguish gravity from inertial accelerations. We’ll
leave the details of that for independent study and get back to accelerometers in the
form of MEMS.
Accelerometer Theory | 415
www.it-ebooks.info
MEMS Accelerometers
Micro-scale accelerometers are not that much different from the machine previously
described but generally use a cantilevered beam instead of a spring. To track more than
one axis, sometimes three discrete accelerometers are placed out of plane with respect
to one another. Alternatively, more complex models use elements that can sense all three
directions within a single integrated sensor. These generally give better results.
The only important difference from the aforementioned examples, besides MEMS being
thousands of times smaller in scale than the mass and spring, is how to measure the
deflection of the test mass. There are three common methods employed in accelerom¬
eters. For most game devices where extreme accuracy isn’t required, the deflection is
usually measured as a change in capacitance. This is somewhat the same way that ca¬
pacitive touch screens work, as described in Chapter 20, and is shown in Figure 21-2.
The beam deflects under the influence of the external accelerations of the test mass and
brings two charged plates farther or closer together. This changes the capacitance of the
system. This change can then be calibrated to the imposed acceleration.
Other methods include integrating a piezo resister in the beam itself so that the deflection
of the beam changes the resistance of the circuit. Although this ultimately gives better
results, these are harder to manufacture. For the most demanding applications, there
are accelerometers using piezoelectric elements based on quartz crystals. These are very
sensitive even during high-frequency changes in acceleration but are generally not used
in sensing human-input motion.
416 | Chapter 21: Accelerometers
www.it-ebooks.info
Common Accelerometer Specifications
To help you better experiment with accelerometers, we’ve collected the specifications
on a few of the most common accelerometers in use at the time of writing. The future
may hold cheap accelerometers based on quantum tunneling that can provide almost
limitless accuracy, but Table 21-1 outlines what you’ll generally be working with for
now.
Table 21-1. Current common accelerometers
1 Device
Accelerometer chip
Sensor range
Sampling rate I
iPhone/iPad/ Motorola Droid
LIS331D
±2g*
100 Hz or 400 Hz
Nintendo Wii
ADXI330
±3 g
x-/y-axis: 0.5 Hz to 1600 Hz
z-axis: 0.5 Hz to 550 Hz
Sony Six Axis
Not published
1+
U)
100 Hz
-
M>?> 4 »
ti t*,
The chip LIS221D is actually capable of two modes. One mode is ±2 g
and the other is ±8g. This is dynamically selectable according to the
chips datasheet; however, neither iOS nor Android allows developers
to change the mode through the API.
The 2 g limit for phones can cause problems when you’re attempting to record motion.
This limitation will be discussed later in this chapter. The larger range of Wii and Sony
controllers demonstrate that they are dedicated to gaming where larger accelerations
are expected.
Data Clipping
The human arm is capable of exceeding the ±2g range of the iPhone’s sensor easily. The
values reported by the API will actually exceed 2 g up to about 2.3 g. The accuracy of
these values that exceed the specification is unknown. Regardless, they are probably at
least as accurate as the option of trying to recreate the data, so if required they can be
used. All values above this upper limit will be reported as the upper limit such that if
you graphed the values, they would look like Figure 21-3.
Accelerometer Theory | 417
www.it-ebooks.info
There are several different ways to handle data clipping. One is to discard the data and
alert the user that he has exceeded the available range. Another is to attempt to recreate
the missing data. If you are recording the data for later processing, you can use both
segment 1 and segment 2 to fit the curve between the point at which the data began to
be clipped and the point in which meaningful data collection is resumed. This is highly
application dependent, and the curve used to fit the data will have to be matched to the
activity at hand. If you are recording the data for later processing, you can use both
segment 1 and segment 2 to give your data.
If you are attempting to process the signal in real time, you’ll have only segment 1 to
work from. This could result in a discontinuity when meaningful data collection re¬
sumes, and you’ll have to decide how to deal with that given the particulars of what you
are doing with the data.
Sensing Orientation
Sensing rotation in three degrees of freedom amounts to sensing a rigid body’s orien¬
tation and is a complex problem that cannot be fully resolved using only accelerometers.
418 | Chapter 21: Accelerometers
www.it-ebooks.info
Think about holding the device vertically. If you rotate the device about the axis de¬
scribed by the gravity vector, none of the accelerometers will measure any change in the
force acting on their test masses. We can’t measure that degree of freedom. To do so,
we’d need to fix a gyroscope to the device, and even these run into problems when a
body is free to rotate about all three axes. See Chapter 11 for a discussion on Euler angles.
Now let’s discuss what we can accomplish. First, Figure 21-4 demonstrates the coordi¬
nate system we’ll use; the actual coordinate system used will be determined by the man¬
ufacturer of your device, so make sure to check its documentation.
♦Y
Figure 21-4. Accelerometer coordinate system
Now if we make some assumptions based on how a user will hold our device, we can
determine some “gross” orientations. For illustration, Table 21-2 gives some idea of what
each value would be for each gross direction, assuming the coordinate system shown
in Figure 21-4.
Table 21-2. Gross acceleration values and orientations
1 Device orientation
X Y
* 1
Facedown on table
0
0
1
Face up on table
0
0
-1
Horizontal on table, right side down
1
0
0
Horizontal on table, left side down
-1
0
0
Vertical on table, bottom down
0
-1
0
Vertical on table, top down
0
1
0
There are a few things to note here. First, if you were to hold the phone in these orien¬
tations with your hand, the accelerometer is sensitive enough to pick up small deviations
Sensing Orientation | 419
www.it-ebooks.info
from true vertical. We are considering these the “gross” orientations such that these
small deviations should be ignored.
Sensing Tilt
Although we can’t determine exactly what angle the user is holding the phone about all
three axes, we can pick one axis, assume that it is pointing down, and then find the
change in the angle from that assumption over time. For instance, if the phone is lying
on a table, the average acceleration in the z-direction will be -1, and in the other di¬
rections, 0. Even if the user spins the phone, the values will remain as previously indi¬
cated and we cannot sense that rotation. However, if the user lifts one edge from the
table—we’ll call this tilting it—then the accelerometer will register different values.
Some of the acceleration due to gravity will act on the other two axes. By sensing this
change, an accelerometer will allow us to determine at what angle the device is tilted.
Using Tilt to Control a Sprite
Here we will show you how to implement code for a simple game that asks the user to
move an avatar to a target by tilting the phone. First, we will briefly show an example
of determining the rotation about a single axis. Let’s assume we have an accelerometer
rotated at some arbitrary angle, a, which is what our algorithm will solve for. As previ¬
ously discussed, accelerometers generally report values as multiples of near earth gravity,
g. For the following example, we are concerned only with the x- and y-axis values, a x
and a y , respectively. If the device were in the “upright” position, then a x would equal 0
and a y would equal 1. After rotating the device, we’d see different values that are related
to our angle a by use of the arctangent function. In this case, because the single-argument
atan function included in most programming languages doesn’t differentiate between
diametrically opposed directions, it is beneficial to use the two-argument function. The
relevant C code is as follows:
tfdefine PI 3.14159
float ftnd2dAngle(votd){
//LOCAL VARIABLES
float alpha,
double ax, ay;
//POLL ACCELEROMETER FOR ACCELERATIONS, API SPECIFIC
ax = getXaccelerationQ;
ay = getYacceleratlon();
//FIND ANGLE
alpha = atan2(ay,ax);
if (alpha >= 0){
420 | Chapter 21: Accelerometers
www.it-ebooks.info
(180/PI);
alpha = alpha *
else {
alpha = alpha * (-180/PI) + 180;
}
return alpha;
}
This is pretty straightforward, but there are a few things to point out. First, the way in
which your program will get results from the accelerometer will vary greatly between
platforms, so we have encapsulated that API-specific code in a getXaccelerationQ
function. In fact, most operating systems will be continuously polling the accelerometer
in a separate thread, so you’ll have to have a logical operator that tells your accelerometer
object when you actually want to see those values passed to your program. Example
Objective-C code for the accelerometer in the iPhone will be shown later. Secondly,
you’ll notice that we are using an if statement that changes the radians to degrees in
such a way as to return proper 0°-360° answers. This avoids having to pay attention to
the sign, as atan2 returns only answers between 0° and 180°, using a negative value to
represent the other half of the range. For example, an output of 0° means the device is
vertical, an output of 90° means the device is rotated 90° to the left, and an output of
180° means the device is upside down.
Now let’s extend this to two dimensions. This will tell us not only how far the phone is
from vertical about one axis, but its inclination about the y-axis as well.
Two Degrees of Freedom
Now let’s say that we want to develop a game in which we control a sprite moving in a
2D world. The user would hold the device as if it were lying on a table and look down
from above. He or she would then tilt the phone out of that plane to get the sprite to
move in the desired direction. The fraction of gravity that the accelerometer is now
experiencing in the x- and y-directions will be inputs into our simulation.
The example will be demonstrated using Objective-C code for the iPhone, and we’ll be
using the Qwartz2D graphics framework. If you aren’t familiar with Objective-C, don’t
worry—we’ll explain what we are doing in each step, and you can port that code to
whatever language you are working in.
The first step will be to set up our accelerometer. In this case we are going to initialize
it in our tiltViewController.m file so that we have:
- (void)viewDidLoad
{
UIAccelerometer *accelerometer = [UIAccelerometer sharedAccelerometer];
accelerometer.delegate = self;
accelerometer.updatelnterval = kPollingRate;
[super viewDidLoad];
}
Sensing Tilt | 421
www.it-ebooks.info
The important concept here is that we have defined a name for our accelerometer object,
accelerometer, and we have set its updatelnterval property to kPollingRate. This
constant was defined in tiltViewController.h as (1.0f/60.0f), which corresponds to 60
Hz. In other words, this tells the operating system to update our program’s accelerometer
object 60 times a second. Also in tiltViewControiler, m, we write what happens when the
accelerometer object gets updated via the accelerometer’s dldAccelerate: function as
follows:
- (void)accelerometer:(UIAcceleropieter *)accelerometer
dldAcceterate:(UIAcceleration *)acceleration{
[(SpriteView *)self.view setAcceleratlon:acceleration];
[(SpriteView *)self.view draw];
}
This function is called every time the acceleration object is updated and does two things.
First, it takes the acceleration data from the accelerometer and passes it to the Sprite
View class, which we’ll talk about in a second. Then it tells the SpriteView to go ahead
and redraw itself.
The SpriteView class is where the action happens and consists of a header file, Sprite
View.h, where we define the following global variables:
UHmage *sprite
A pointer to the image that will be used to represent our sprite on the screen.
currentPos
The position on the screen where we want the sprite to be drawn.
prevPos
The previous position of the sprite on the screen. We will use this to tell the draw
function what parts of the screen need to be redrawn.
UIAcceleration *acceleration
A special Objective-C data type to hold data from the accelerometer.
CGFloat xVelocity and CGFloat yVelocity
Float variables to hold the current velocity in the x-direction and y-direction, re¬
spectively.
CGFloat convertX and CGFloat convertY
Float variables to hold the ratios for converting our physics engine’s results in meters
to pixels based on an assumed world size.
Additionally, we’ve defined the following global constants:
9
Near earth gravity value, set at 9.8 m/s 2 . This will convert the accelerometer’s values
from g to m/s 2 for use in calculating velocity. This can also be tuned to represent an
422 | Chapter 21: Accelerometers
www.it-ebooks.info
arbitrary acceleration instead of just using gravity as the force (e.g., percent of jet
engine thrust).
kWorldHeight and kWorldWidth
These values are used to allow the programmer to change the assumed world di¬
mensions. Higher values mean each pixel is a greater distance in meters. The world
will always be scaled to fit on the screen, so a large world means the sprite will appear
to move slower (a few pixels at a time) for a given acceleration. Note that our current
code doesn’t scale the sprite.
Now we’ll show you how we use these variables in SpriteView. m to move our sprite on
our screen as a result of the accelerometer values. First, we have some initialization to
do, which takes place in the initWithCoder: method that runs the first time the view
is loaded:
-(id)initWithCoder:(NSCoder *)coder {
if((setf = [super inttWithCoder:coder])){
self.sprite = [Ullmage imageNapied:@"sprite.png"];
self.currentPos = CGPointMake((self.bounds.size.width / 2.Of) +
(sprite.size.width / 2.0f), (self.bounds.size.height /2.0f)+(sprite.size.height /2.Of));
xVelocity = 0.0f;
yVelcoity = 0.0f;
convertX = self.bounds.size.width / kWorldWidth;
convertY = self.bounds.size.height / kWorldHeight;
}
return self;
}
Most of this is pretty straightforward. We tell our program where to find the sprite image
we’ve chosen and set its initial position to the center of the screen. We also set its initial
velocity to 0 in both directions. We then go ahead and initialize our convertX and
convertY variables based on the self. bounds. size property, which gives the bounds
of the view in pixels. We’ll show exactly how this affects our program later. Next, we’ll
write a custom mutator for the CurrentPos variable:
- (vold)setCurrentPos:(CGPoint)newPos {
prevPos = currentPos;
currentPos = newPos;
lf(currentPos.x <0){
currentPos.x = 0;
xVelocity = 0.0f;
}
lf(currentPos.y <0){
currentPos.y = 0;
yVelcoity = 0.0f;
}
lf(currentPos.x > self.bounds.size.width - sprite.size.width){
Sensing Tilt | 423
www.it-ebooks.info
currentPos.x = self.bounds.size.width - sprite.size.width;
xVelocity = 0.0f;
}
if(currentPos.y > self.bounds.size.height - sprite.size.height){
currentPos.y = self.bounds.size.height - sprite.size.height;
yVelocity = 0.0f;
}
CGRect curSpriteRect = CGRectMake(currentPos.x, currentPos.y,
currentPos.x+sprite.size.width, currentPos.y+sprite.size.height);
CGRect prevSpriteRect = CGRectMake(prevPos.x, prevPos.y,
prevPos.x+sprite.size.width, currentPos.y+sprite.size.height);
[self setNeedsDisplaylnRect:CGRectUnion(curSpriteRect, prevSpriteRect)];
}
In case you are unfamiliar with Objective-C, when you define a class instance variable
it will automatically define a mutator that simply updates the value of the variable to
the value you are passing it. However, in the preceding example we are overriding that
mutator to do some additional work. The first thing we do is to set the prevPos variable
to the current position of the sprite and then update the currentPos with the value the
mutator was given. However, our physics engine isn’t going to include collision response
with the screen boundaries, so we go on to check if the sprite has reached the screen
edge. If so, we simply tell the program to leave it on the edge and to set the velocity in
that direction to 0. Lastly, we define a couple of rectangles based on the new position of
the sprite and the old position of the sprite. After we union those rectangles together,
we tell the operating system to redraw the screen in that area with the setNeedDis
playlnRect: method. As you might recall, our accelerometer object is calling the draw
method every time it updates, and it is in this method that we will put our physics engine:
- (vold)draw {
static NSDate *lastllpdateTime;
if (lastUpdateTine != nil) {
NSTlmelnterval secondsSlncetlpdate = - ([lastUpdateTine
tlmelntervalSlnceNow]); //calculates Interval In seconds from last update
//Calculate displacement
CGFloat deltaX = xVelocity * secondsSlncetlpdate +
((acceleration .x*g*secondsSlncetlpdate*secondsSlncetlpdate)/2); // METERS
CGFloat deltaY = yVelocity * secondsSlncetlpdate +
((acceleration .y*g*secondsSlncetlpdate*secondsSlncetlpdate)/2); // METERS
//Converts from meters to pixels based on defined World size
deltaX = deltaX * convertX;
deltaY = deltaY * convertY;
//Calculate new velocity at new current position
xVelocity = xVelocity + acceleration.x * g * secondsSlncetlpdate; //assumes
acceleration was constant over last update Interval
424 | Chapter 21: Accelerometers
www.it-ebooks.info
yVelocity = yVelocity - (acceleration.y * g * secondsSinceUpdate); //assumes
acceleration was constant over last update interval
//Mutate currentPos which will update screen
self.currentPos = CGPointMake(self.currentPos.x + deltaX,
self.currentPos.y + deltaY);
}
[lastUpdateTime release];
lastllpdateTime = [[NSDate alloc] init];
}
Previously, we discussed issues with timing when working with accelerometer data. In
this case, Objective-C makes it very easy to get the correct elapsed time in seconds. We
first define a static variable, lastUpdateTime, as an NSDate type. This type has a built-
in function to give the time interval in seconds from now, which we assign to an NSTime
Interval variable. Skipping down to the last two lines, we are simply updating the last
update time by releasing and reinitializing the variable. As it is static, it will remain even
after the function returns. If you are using a lower-level language, you might have to
write your own timelntervalSinceNow function that takes into account the particular
clock frequency of the system.
Now that we have our time interval in seconds, we can calculate our new position. Recall
from Chapter 2:
S 2 = si + vi t + (a t 2 )/2
which we have rearranged to be:
As = S 2 - si = vj t + (a t 2 )/2
This gets programmed as:
CGFloat deltaX = xVelocity * secondsSinceUpdate +
((acceleration.x*g*secondsSinceUpdate*secondsSinceUpdate)/2); // METERS
We then convert this displacement in meters to displacement in pixels using an appro¬
priate ratio for the size of our world. Before we can move on, we have to calculate our
new velocity at our new position. So we again assume the acceleration as constant over
the update interval, and recalling:
v 2 — v i + a At
from Chapter 2, we can solve for the new xVelocity with:
xVelocity = xVelocity + acceleration.x * g * secondsSinceUpdate;
Sensing Tilt | 425
www.it-ebooks.info
As you can see from the complete method description, the code of the y-direction is
similar. Finally, we call the currentPos mutator to set the new position based on the
change in displacements. Recall that this is a custom mutator that also tells the operating
system to update the display. After the draw method is finished, the accelerometer waits
1/60 of a second and then calls it again. You could extend this program by adding in
friction, fluid resistance, and collisions with the screen boundaries using the methods
outlined in the other chapters of this book.
426 | Chapter 21: Accelerometers
www.it-ebooks.info
CHAPTER 22
Gaming from One Place to Another
Once a tool meant to help the United States guide intercontinental ballistic missiles, the
Global Positioning System (GPS) has evolved to be a part of our everyday lives. The
current generation will never have known a world where getting lost was something
that couldn’t be fixed by trilaterating their position between satellites orbiting the planet.
Although GPS has become commonplace in the navigational world, the proliferation
of smartphones is just now opening the doors to GPS gaming. While this genre is just
emerging, we’d like to give you an introduction to the physics behind GPS and the
current applications in the gaming world.
Let’s recall that positions near the earth’s surface are generally given in the geographic
coordinate system, more often described as latitude, longitude, and altitude. Latitude is
a measure in degrees of how far north or south you are from the equator. Longitude is
the measure in degrees of how far east or west you are from the prime meridian. A
meridian is a line of constant latitude that runs from the North Pole to the South Pole.
The prime meridian is arbitrarily defined as the meridian that passes through the
Greenwich Observatory in the UK. Altitude is usually given as the measure of how far
above or below sea level you are at the point described by latitude and longitude.
Location-Based Gaming
Before getting to the physics behind GPS, we’d like to take a moment to discuss how
GPS is being implemented into games. Right now, this is an emerging market that is
just starting to gain traction. There are several broad categories into which games fall.
Another step beyond what the accelerometer did, GPS enables users to move computer
games not only off the couch but also out into the world.
427
www.it-ebooks.info
Geocaching and Reverse Geocaching
Geocaching is the oldest form of gaming involving GPS. It originated after selective
availability was removed from GPS, making it more accurate, in the year 2000. In its
most basic form, it is the process of hunting down a “cache” using provided GPS coor¬
dinates. The cache usually has a logbook and may contain other items such as coins
with serial numbers that the finder can move to another cache and track online.
Because of the large amount of setup involved in implementing a geocaching game on
a commercial scale, most implementations are community based. However, reverse ge¬
ocaching has more promise for the gaming industry. In this variation there is nothing
at the supplied coordinates, but traveling to them is required to execute some action.
Think of it as carrying around a cache that cannot be unlocked until it is within range
of some specific coordinate. This could be used to force users to travel in order to unlock
a game item. For instance, perhaps to gain the ability to use a sword in a game, the user
must travel to the nearest sporting goods store. The commercial possibility of corporate
tie-ins is an obvious plus.
Mixed Reality
Mixed-reality games are similar to geocaching. They go beyond just using the coordi¬
nates of the user to trigger events, to using reality-based locals. A current example is
Gbanga’s Famiglia. In this game your movement in the real world allows you to discover
virtual establishments in the game world. This divorces it from the actual physical lo¬
cations that your GPS is reporting but requires moving between locations in the real
world to move your character in the virtual world. Popular right now is the Foursquare
app on mobile devices. This is the simplest possible implementation of mixed-reality
gaming. Foursquare allows a user to become the mayor of a place if she “checks in” at
the locale more than anyone else.
Street Games
Street games are another step beyond mixed reality. These turn the environment around
the user into a virtual game board. One example is the recent Pac-Manhattan multi¬
player game using GPS in smartphones to play a live version of Pac-Man in Washington
Square Park. In general, the idea is to create a court for game play using the environment
surrounding the user. The relationship between users is tracked in the virtual space of
the game and provides the interactive elements.
428 | Chapter 22: Gaming from One Place to Another
www.it-ebooks.info
What Time Is It?
The story of GPS really begins with a prize offered by the British government in 1717
for a simple way to determine your longitude. Awarded in 1773, the accepted solution
was to compare local noon to the official noon sighted at the Greenwich Observatory.
The difference between these two times would allow you to tell how far around the
world you were from the observatory. Fast-forward three centuries, and we have satel¬
lites orbiting the earth, broadcasting time-stamped messages. By calculating the differ¬
ence between the time the message was received and the time it was transmitted, we can
calculate our distance from the satellite. In both cases you need an accurate way to keep
or measure time. For sailors in the 1800s, the device was the newly invented chronom¬
eter. For us, it is the atomic clock.
Because the signals from a GPS satellite are moving at the speed of light, you need a very
accurate clock to keep track of how long it took to travel to you. For instance, if the clock
you are using to time when the signal arrives is 1 microsecond off, you will estimate a
distance over 900 miles in error. On the supply side of the signal, each satellite has an
atomic clock, and internal GPS time is accurate to about 14 nanoseconds. The problem
is that you also need a very accurate clock in the receiver, and it would be pretty hard
to fit an atomic clock into a phone economically. To get around this, the receiver must
figure out the correct current time based on the signals from the satellites.
Two-Dimensional Mathematical Treatment
This section will give you a good idea of how GPS systems determine their location.
This background will help you in many applications of geometry in games in general,
but most GPS devices do the heavy lifting and report through an API your current
latitude and longitude. Some APIs may include more information—for example, the
current iOS API, called Core Location, gives the current latitude and longitude, the
direction of travel, the distance traveled, and the distance in meters to a given coordinate.
It also gives an estimate for the error associated with its position fix in meters.
One way to get your position via the kind of information that GPS provides is a technique
called trilateration. We are going to give this problem a mathematical treatment in two
dimensions. You could extend this to three dimensions by using spheres instead of
circles.
To begin, we can list our unknowns: our x coordinate and y coordinate in space, and
the error in our receiver’s clock (or bias), b. In a two-dimensional plane, trilateration
among three circles gives you an exact position; in three-dimensional space, four
spheres are required to determine all three special coordinates. Note that if we included
an assumption about being on the surface of some geometric shape, such as the earth,
What Time Is It? | 429
www.it-ebooks.info
we could reduce the number of unknowns. No such simplification is used here to pro¬
vide you with the most general case.
In our example, we are somewhere on the surface of the two-dimensional earth, shown
in Figure 22-1 as a light gray solid disk. This diskis being orbited by several GPS satellites.
The satellites’ orbits are regular, and their positions at any time are tabulated in an
almanac that is stored in the receiver. The time of transmission is encoded in the signal
so that the givens are x t , y„ and f, with i =1,2,3.
To make things easier for us, we are going to abandon the coordinate system of the earth
and use the coordinate system defined by our three satellites. The origin will be at satellite
1, the x-axis going directly from satellite 1 straight to satellite 2 and the y-axis being
perpendicular to that. This is shown in Figure 22-2.
430 | Chapter 22: Gaming from One Place to Another
www.it-ebooks.info
The equations of the three circles are therefore:
9 9 9
r i = x“+y z
r 2 2 = (x-d) 2 +y 2
r 2 3 = (x-i) 2 +(y-j) 2
Each radius is found by subtracting the transmission time, f, from the current time and
multiplying by the speed of light. As the speed of light is very large, and our current
time is only a rough estimate, these radii are commonly referred to as pseudoranges to
remind us they are still approximate. The x and y values that satisfy these equations are
our current location in two-dimensional space. We now subtract the second equation
from the first:
r 1 2 -r 2 2 = x 2 + y 2 -(x-d) 2 -y 2
rj 2 - r 2 2 = x 2 - (x-d) 2
Solving for x gives:
x = ( ri 2 - r 2 2 +d 2 ) / 2d
What Time Is It? | 431
www.it-ebooks.info
where d is the distance between the known locations of satellite 1 and satellite 2. We
now substitute our x coordinate back into the first circle’s equation:
y 2 = rj 2 _ [(( rj 2 _ p,2 + d 2)2) / ( 4d 2)]
and finally after we take the square root:
y
^ V' 1
r 2 + d 2 f
4 d 2
notice that they value is now expressed as a positive or negative square root. This means
there can be zero, one, or two real-number solutions. If the circles do not intersect, then
the quantity under the square root will be negative and the y value will have zero real
solutions. This is unlikely for the first two satellites because you have already received
pseudoranges to them in the form of r 1 and r 2 , which the algorithm assumes to have
zero error. If the two circles happen to intersect at only a single tangent point, they will
have one solution and will be equal to 0. This is also unlikely. The most likely result will
be that y will be the set of two values, plus or minus the square root of a positive value,
and that those two points will be widely separated.
Now if we included the assumption that we were on the earth’s surface, we could already
break the tie between the two points by picking whichever was closest to the earth’s
surface. However, we still would have to deal with the likelihood that there is a large
error in our position given the imprecise clock in the receiver.
We can fix our position ( x,y) with no assumptions and account for clock bias by intro¬
ducing the third point and its pseudorange. Now, it was very likely that the circles
obtained from the first two satellites would intersect because of the way that GPS sat¬
ellites are arranged around the earth. However, because our calculations use pseudor¬
anges, it is relatively unlikely that the third circle will pass directly through one of the
two points defined by the intersection of the first two circles. To remove such clock-
related distance errors, we first calculate which point (x,y) or (x,-y) is closer to ( i,j ) and
choose that to be our assumed location. The difference between the smaller of these two
distances and the pseudorange r 3 is then our distance correction, d a . As the signal is
traveling at the speed of light, the following quotient provides an estimate of the error
between the correct time and the receiver’s time:
b = d a /c
As all the GPS satellites have synchronized atomic clocks, the same bias exists for each.
This means that the bias we calculated would actually affect the first pseudoranges we
used to find our initial bias. Therefore an iterative approach is required to adjust all the
variables in real time until they converge. A more direct, but less obvious, algebraic
432 | Chapter 22: Gaming from One Place to Another
www.it-ebooks.info
solution that requires no iteration was developed by Stephen Bancroft. It is detailed in
his paper “An Algebraic Solution of the GPS Equations” in the IEEE Transactions on
Aerospace and Electronics Systems journal.
Besides clock errors, other errors are introduced by the atmosphere, signals bouncing
off the ground and back to the receiver, relativistic effects (discussed in Chapter 2), and
atomic clock drift. These are all accounted for in mathematical models applied to the
raw position data. For instance, the GPS clocks lose about 7,214 nanoseconds every day
due to their velocity according to special relativity. However, because they are higher
up in the earth’s gravity well, they gain 45,850 nanoseconds every day according to
general relativity. The net effect is found by adding these values together: they run 38,640
nanoseconds faster each day, which would cause about 10 kilometers inaccuracy to build
each day they are in orbit. To account for this, the clocks in the GPS receivers are pre¬
adjusted from 10.23 MHz to 10.22999999543 MHz. The fact that we are giving you a
number to 11 decimal places demonstrates the amount of accuracy the modern age
enjoys in its time keeping.
Once the bias is taken care of and all the other possible errors adjusted for, the converged
solution can be translated back into whatever coordinate system is convenient to give
to the end user. Usually this is latitude, longitude, and altitude. Next, we will learn how
to calculate different quantities based in the geographic coordinate system.
Location, Location, Location
Let’s take a minute to discuss distance between two latitude and longitude coordinates.
You might be tempted to calculate it as the distance between two points. For very small
distances, this approximation is probably accurate enough. However, because the earth
is actually a sphere, over great distances the calculated route will be much shorter than
the actual distance along the surface.
The shortest distance between two points on a sphere, especially in problems of navi¬
gation, is called a great circle. A great circle is the intersection of a sphere and a plane
defined by the center point of the sphere, the origin, and the destination. The resulting
course actually has a heading that constantly changes. On ships, this is avoided in favor
of using a rhumb line, which is the shortest path of constant heading. This makes nav¬
igation easier at the expense of time. Airplanes, however, do follow great-circle routes
to minimize fuel burn.
Distance
There are several ways of calculating the distance along a great circle. The one we will
discuss here is the haversine formula. There are other methods like the spherical law of
cosines and the Vincenty formula, but the haversine is more accurate for small distances
Location, Location, Location | 433
www.it-ebooks.info
than the spherical law of cosines while remaining much simpler than the Vincenty
formula.
The haversine formula for distance is:
d = (R)(c)
where R = earths radius and c is the angular distance in radians given by:
c — 2 arcsin (V«)
Here, a is the square of half the chord length between the two points, calculated as:
a = sin 2 (Ai at /2) + cos(lati)cos(lat 2 )sin 2 (Ai ong IT)
Finally:
A^g = long 2 - long!
Ajat = lat 2 - lat!
Remember to first convert the angles to radians before using them in the trig function.
Next we will begin showing you an implementation of several different formulas in
Objective-C; however, these should translate to C with little modification. These all use
the following data structure to hold latitude and longitude information:
typedef enum {
float lat;
float Ion;
} Coordinate2D;
Given this, the haversine implementation would look like:
float dlstanceGreatClrcle(Coordlnate2D startPoint, Coordinate2D endPolnt){
//Convert location from degrees to radians
float latl = (M_PI/180.) * startPoint.lat;
float lonl = (M_PI/180.) * endPoint.longi;
float lat2 = (M_PI/180.) * endPoint.lat;
float lon2 = (M_PI/180.) * endPoint.longi;
//Calculate deltas
float dLat = lat2 - latl;
float dLon = lon2 - lonl;
//Calculate half chord legnth
float a = sin(dLat/2) * sin(dLat/2) + cos(latl) * cos(lat2) * sin(dLon/2)
* sin(dLon/2);
434 | Chapter 22: Gaming from One Place to Another
www.it-ebooks.info
//Calculate angular distance
float C = 2 * atan(sqrt(a)/sqrt(l-a));
//Find arclength
float distance = 6371 * C; //6371 Is radius of earth in kn
return distance;
}
One limitation of the preceding method is that if the two locations are nearly antipodal
—that is, on opposite sides of the earth—then the haversine formula may have round¬
off issues that could results in errors on the order of 2 km. These, however, will be over
a distance of20,000 km. If extreme accuracy is required for nearly antipodal coordinates,
you can fall back to the spherical law of cosines, which is best suited for large distances
such as the antipodal case.
Great-Circle Heading
As discussed before, to follow the shortest path between two points on a sphere you
must travel along a great circle. However, this requires that your heading be constantly
changing with time. The formula to calculate your initial heading, or forward azi¬
muth, is:
©i = atan2[sin(Ai ong )cos(lat2), cos(lati)sin(lat 2 ) -
sin(lati)cos(lat 2 )cos(Ai ong )]
Recall that atan2 is the two-argument variation of the arctangent function. It returns
a normalized angle in radians between -tt and tt (-180° and 180°). The code that cal¬
culates the value and returns the compass bearing is as follows:
float initialBearing (Coordinate2D startPoint, Coordinate2D endPoint){
//Convert location from degrees to radians
float latl = (M_PI/180.) * startPoint.lat;
float lonl = (M_PI/180.) * startPoint.longi;
float lat2 = (M_PI/180.) * endPoint.lat;
float lon2 = (M_PI/180.) * endPoint.longi;
//Calculate deltas
float dLat = lat2 - latl;
float dLon = lon2 - lonl;
// Calculate bearing in radians
float theta = atan2f( sin(dlon) * cos(lat2), cos(latl)*sin(lat2)-sin(latl)*cos(lat2)
*cos(dlon));
//Convert to compass bearing
Float bearing = theta * (180 / M_PI); //radians to degrees
bearing = ( bearing > 0 ? bearing : (360.0 + bearing)); //fix range
return bearing;
}
Location, Location, Location | 435
www.it-ebooks.info
A negative angle involves starting at 0° and rotating in the decreasing-heading direction,
but compasses aren’t labeled with negative values! To fix this, the line that has the com¬
ment “fix range” is using a ternary operator to say that if the bearing is less than 0, return
the value the compass would read. For example, if the bearing were -10°, then the
compass bearing is -10° + 360° = 350°. If the value is positive, then it just returns the
same value.
To find the final bearing, we simply take the initial bearing going from the end point to
the start point and then reverse it. The code is produced as follows:
float flnalBeartng (Coordlnate2D startPoint, Coordinate2D endPoint){
//Convert location from degrees to radians
float latl = (M_PI/180.) * endPoint.lat;
float lonl = (M_PI/180.) * endPoint.longi;
float lat2 = (M_PI/180.) * startPoint.lat;
float lon2 = (M_PI/180.) * startPoint.longi;
//Calculate deltas
float dLat = lat2 - latl;
float dLon = lon2 - lonl;
//Calculate bearing in radians
float theta = atan2f( sin(dlon) * cos(lat2), cos(latl)*sin(lat2)-sin(latl)*cos(lat2)
*cos(dlon));
//Convert to compass bearing
float bearing = theta * (180 / M_PI); //radians to degrees
bearing = ( bearing > 0 ? bearing : (360.0 + bearing)); //fix range
bearing = ((bearing + 180) % 360) //reverse heading
return bearing;
}
The difference here is that we have flipped latl, longi and lat2, long2 while converting
the locations to radians. Also, before we return the bearing value, we reverse it by adding
180° degrees to it. The modulo operator (%) ensures that values over 360° are rolled over
into compass coordinates. For example, if we calculate a bearing of 350° and add 180°
to it, we get 530° degrees. If you start at 0° and go around 530°, you’ll end up at 170°.
The modulo operator will result in the bearing being calculated with this correct com¬
pass value.
Rhumb Line
As discussed before, it is sometimes preferable to take a longer path of constant heading,
called a rhumb line, as compared to constantly changing your heading to follow a great
circle path. The rhumb line will be longer than the great circle, and the distance you are
from the great circle route at any moment is called the cross track error. To cross the
Atlantic is about 5% longer if you follow a rhumb line. The extreme example of going
from the East Coast of the United States to China is about 30% longer. However, such
436 | Chapter 22: Gaming from One Place to Another
www.it-ebooks.info
large penalties are rarely encountered because ships have to alter course to avoid land!
This makes “as the crow flies” examples unrealistic.
If your game is providing navigation information to anyone but pilots, it will probably
be using rhumb lines. The following are the formulas used to calculate distance and
bearing between two coordinates on a rhumb line. The easiest way to begin is to flatten
the globe. In a Mercator projection, rhumb lines are straight. In fact, this makes graph¬
ically solving the problem very simple. You use a ruler. Mathematically, things get a bit
more complicated. The following equation gives Acp, which is the difference in latitude
after taking into account that we have stretched them in order to flatten the sphere:
Acp = ln[tan(lat2/2 + jt/4) / tan(latl/2 + tc/4)]
The distance between two points on a rhumb line is given by:
D — ^Alat + q 2 * Allng * R
The variable q is a value whose formula depends on Acp. If Acp is equal to 0, that means
that the calculated course is going to be either directly east or west. If that is the case,
then the intermediate value of q is:
q = cos(latl)
if Acp is not equal to 0, then:
4 = Ai a t/Acp
You can see that if not properly implemented, a direct east or west course would result
in division by 0. Finally, the constant bearing is:
®rhumb — a.tan2(A] on g, Acp)
There are, in fact, an infinite number of rhumb lines that will get us to our end point.
However, the longer ones will either take us the wrong way around the globe, or spiral
around the globe before hitting our end point. At any rate, the shortest route will be the
one in which A long is less that 180°. The preceding is implemented in Objective-C as
follows:
float rhumbBearing ( Coordinate2D startPolnt, Coordinate2D endPoint){
//Convert location from degrees to radians
float latl = (M_PI/180.) * startPoint.lat;
float lonl = (M_PI/180.) * startPoint.longi;
float lat2 = (M_PI/180.) * endPoint.lat;
float lon2 = (M_PI/180.) * endPoint.longi;
Location, Location, Location | 437
www.it-ebooks.info
//Calculate deltas
float dLat = lat2 - latl;
float dLon = lon2 - lonl;
//find delta phi
float deltaPhi = log(tan(lat2/2+(M_PI)/4)/tan(latl+M_PI/4))
float q=(deltaPhi==0 ? dlat/deltaPhi : cos(latl); //avoids division by 0
if (abs(dLon) > M_PI){
dLon = (dLon>0 ? -(2*(H_PI-dLon):(2*M_PI+dLon));
}
float D = sqrt(dLat*dLat + q*q*dLon*dLo)* 6371;
float theta = atan2f(dLon, deltaPhi);
//now convert to compass heading
float bearing = theta * (180 / M_PI); //radians to degrees
bearing = ( bearing > 0 ? bearing : (360.0 + bearing)); //fix range
return bearing;
}
There are a few things worth pointing out. First is that we are using a ternary function
in the line commented by “avoids division by 0” to take care of the case when delta
Phi is equal to 0. If it is 0, q is set to cos(latl); if not, then it is set to dlat/deltaPhi.
The If statement immediately following ensures that if dLon is greater than it (180°),
hence putting us on a longer-than-required rhumb line, then we should correct the
value to correspond to the shortest route. This is achieved via the ternary, which ensures
that dLon is less than it and nonnegative. Lastly, we convert from a normalized radian
answer to a compass direction.
Now that you have a good idea about how to calculate position and distance in the
geographic coordinate system, you can use the earlier chapters to determine other
quantities like speed and acceleration.
438 | Chapter 22: Gaming from One Place to Another
www.it-ebooks.info
CHAPTER 23
Pressure Sensors and Load Cells
Pressure sensors are an evolution of the simple button. A simple button has two states,
on or off, which can be used to trigger simple atomic actions in a video game such as
firing a gun or opening a door. However, simple buttons are not capable of informing
the program how you, the user, hits that button. Did you hit it quickly? Did you barely
touch it at all? The only thing the program can interpret is that you did in fact hit the
button.
With pressure sensors, the program has the ability to discern how the user pressed the
button. This information can be used as incremental input, such as the player raising a
firearm before pressing the button harder to actually fire. Additionally, pressure sensors
can be used to create novel forms of human-input devices. While pressure sensitivity is
not uncommon in the more traditional console gaming markets, there is also a recent
push to move the sensors into touch-screen devices like the Nintendo DS and cell phone
gaming market. Pressure-sensitive touch screens are currently beyond state of the art,
however, so we’ll primarily discuss the traditional methods already in widespread adop¬
tion.
In addition to pressure sensors, some new gaming consoles use load cells to allow the
player to use shifts in his or her body weight as input. The method by which this data
is collected and how the center of gravity is determined will be discussed in this chapter.
Lastly, some smartphones now include a barometer, a pressure sensor that measures the
pressure of the atmosphere. What it is used for and the type of information it can provide
will also be discussed.
439
www.it-ebooks.info
Under Pressure
As discussed in Chapter 3, pressure is a force applied over an area. Imagine a concrete
block sitting on a steel plate. The weight of the block will be evenly distributed over the
area of contact, creating a pressure on the steel plate. Gas and liquid can apply pressure
as well. The weight of the air pressing down on us is what is known as atmospheric
pressure.
Let’s cover a quick example of how to calculate pressure just to illustrate the concepts
involved. Pressure has many different units, but all of them can be equated to a force
divided by an area. For this chapter we’ll stick with Newtons per square meter, as this
is easiest to visualize. The SI derived unit (a unit of measure made up of other funda¬
mental units) is called a Pascal, which is just 1 N/m 2 .
Example Effects of High Pressure
In Chapter 3, we discussed the concept of buoyancy and how it arises from hydrostatic
pressure. Here, we’ll show the tremendous forces that hydrostatic pressure can cause
on a submerged object. Let’s imagine we have a steel ball filled with normal atmospheric
pressure at sea level, or about 101,000 N/m 2 . While this seems like a lot, your body is
used to dealing with this pressure, so you don’t even notice it on a daily basis! Now we
are going to take this ball and drop it into the Marianas trench, the deepest known part
of the ocean. The water depth here is approximately 10,900 meters. The formula for
calculating the pressure due to water (hydrostatic pressure) is:
P(h) = p X g X h
where p is the mass density of water, g is force due to gravity, and h is the height of the
water column above the object.
Here we take the standard density for saltwater, 1025 kg/m 3 , and calculate what the
pressure is:
P(10,900) = (1025 kg/m 3 ) x (9.8 m/s 2 ) X (10,900 m) =
109,490,500 N/m 2
Figure 23-1 shows how the pressures act against our steel ball.
440 | Chapter 23: Pressure Sensors and Load Cells
www.it-ebooks.info
/ / --—""
*Y Air
r pressure /
4 r~
Hydrostatic pressure
Figure 23-1. Pressure differential
It is clear that the water pressure acting on the sphere is much larger than the air pressure
we trapped inside before sinking it. Also, note that pressure always acts normal to the
surface. If you happen to apply a force to the vertex of an object, you’ll have trouble
modeling the right effect because a vertex does not have a well-defined normal. We can
overcome this only by applying pressure to the faces of polygons or by averaging the
direction of the pressure on either side of the vertex. Returning to our example, the net
pressure differential on the steel ball is:
P(water) - P(air) = 10,949,100 N/m 2 - 101,000 N/m 2 =
10,848,100 N/m 2
This is the pressure you would feel if 1,870 elephants were standing on a 1 -square-meter
plate on top of you. If our steel ball had walls that were too thin to withstand this pressure
differential, it would implode. To put this all into perspective, a steel ball in space would
have a pressure differential of only 1 atmosphere pushing out. It is thus much harder to
design a structure to go to the bottom of the sea than it is to go to the moon. Indeed,
more men have stepped foot on the lunar surface than have visited the bottom of the
sea.
Under Pressure | 441
www.it-ebooks.info
If the ball were open to the sea, then the pressure would act equally on each side of the
steel boundary. Without a pressure differential, there would be no force to crush the
ball; however, there would still be compression of the steel shell itself.
Button Mashing
While the preceding example highlights some important concepts about pressure in
general, it is not usually the type of pressure used as input to a game. The most common
types of pressure sensors you’ll experience in video games are pressure-sensitive buttons
that indirectly measure the amount of pressure the user is exerting on the button and
convert this to a relative value. Both Sony and Microsoft have incorporated pressure-
sensitive (also known as analog) buttons into their controllers for the PlayStation and
Xbox series of consoles.
The method by which you can detect how hard a user is pressing a button varies from
very simple to very complex. We’ll focus on Sony’s method, which is very elegant. A
typical push button is just two contacts separated by an insulator, most commonly air.
When the button is pushed, the upper contact moves down and touches the lower
contact. This completes a circuit, causing a voltage spike, which the device interprets as
a button press. This is another example of a digital sensor—it is either on or off. The
buttons in Sony’s controller work a bit differently. In State A in Figure 23-2, we can see
that the button is not yet pressed and an air gap exists between the solid conductor and
the domed flexible conductor. In State B the button is depressed with minimal pressure,
and the dome just barely makes contact. The button is now activated. If the user con¬
tinues to press down harder on the button, the dome deflects and increases the area of
contact with the fixed conductor; the larger the contact area, the greater the conductivity
of the connection. This causes a rise in the current flowing in the circuit.
442 | Chapter 23: Pressure Sensors and Load Cells
www.it-ebooks.info
By measuring this increase in current, the controller knows how far down the button
is being pressed. In State D in Figure 23-2, the button is at its limit of travel and the
dome has deflected to its maximum contact area. The difference between this maximum
and the minimum required to detect contact determines the absolute lowest and highest
pressure the button is able to differentiate. For instance, let us assume that if the button
were depressed completely, the current would register at f max . If the button were not
pressed at all, of course, the current would be 0. If we call the current J(f) for any time,
t, we see that the ratio f(f)/I max gives a nondimensional quantity for how far down the
button is pressed. During this operation, the hardware converts the analog voltage to a
digital representation suitable for input to a program. For the Sony example, this value
is calculated by the hardware and passed as part of the data stream from the controller
with hex values between 0x00 to OxFF, or in other words, integers 0 to 255 in decimal.
This means that each buttons travel is divided into 255 parts that your program can
register.
While 255 individual increments are beyond the human ability to control fingertip
pressure, different ranges of pressure have practical uses in games. For example, you
could program your button to raise a weapon with a half-press (0 to 127), bring the
weapon to the shoulder with more pressure (127 to 250), and to fire when totally de¬
pressed (250 to 255). Of course, those values would have to be tuned for the desired
Button Mashing | 443
www.it-ebooks.info
level of sensitivity. Another example would be to control the throttle on a car by using
the values of 0 to 255 as thrust multipliers.
Another use of knowing a buttons position would be tracking it over time. With a time
history of position, you can differentiate to get velocity and again to get acceleration.
This would allow the program to differentiate between a button that is either slowly
depressed or quickly depressed. Most hardware doesn’t help you here, so you’ll have to
store the values and calculate the velocities in whatever increments are appropriate for
your program. As real-time velocity sensing might be taxing to the user as real-time
input, the best use would be as input to something that the user doesn’t have to control
constantly. Imagine having to keep a button pressed down at the correct pressure for
your gameplay for longer than a few minutes; I can feel my wrist cramping now. How¬
ever, the pressure button is useful for many inputs. For example, how far a button is
pressed down might be used to draw back the head of a putter, while the speed at which
the button is released could be used to determine the speed at which the putter is brought
back to the ball.
Load Cells
Beyond simple buttons, there are other novel ways to use pressure sensors to allow a
user to interact with your games. For example, Nintendo’s Wii uses a balance board
peripheral based on load cells to detect a person’s stance.
a *
r»«y—
Ms
M>?’ 4 «
71 a*
Tiny scales
Load cells work differently than the pressure-sensitive buttons described previously, but
like pressure-sensitive buttons, they come in different types, all of which measure the
load pressing on them. The most common way, and the one used in the Nintendo
balance board we’ll discuss shortly, is through what is called a strain gauge.
A strain gauge, as you might be able to guess, does not measure force directly but instead
measures how much strain the gauge is experiencing. Strain is a measure of how much
a rigid body has deformed independent of its rigid-body motion. While there are several
notions of strain in continuum mechanics, the one we are concerned with here is often
referred to as engineering strain. This type of strain quantifies how much a structural
element has deformed compared to its original, or rest, length. We normalize this by
dividing the change in length over the rest length. By testing the material, one can
develop a stress versus strain curve that relates how much stress it takes to cause a certain
The original idea for the Nintendo balance board came to video game
designer Shigeru Miyamoto after he was inspired by watching sumo
wrestlers weigh themselves with each leg on a different scale. They are
too heavy to use one scale!
444 | Chapter 23: Pressure Sensors and Load Cells
www.it-ebooks.info
amount of strain. Once the pressure that causes an amount of strain is known, it is
possible to determine the amount of load. Now you might be wondering how the strain
gauge measures the amount that the Wii’s legs compress when you stand on them.
One of the most common electronic strain gauges is th e piezoresistive strain gauge. The
simplest example of a piezoresistive strain gauge would be a single wire. If you were to
elongate a wire from its rest length, the cross-sectional area decreases. This causes a rise
in the electrical resistance of the wire. After measuring the rest resistance, you can use
the difference to determine how much the wire has elongated. Knowing the mechanical
properties of the wire, you can also determine how much force it takes to stretch the
wire.
To make strain gauges sensitive without having long linear wire elements, the conductive
material is often arranged in a strain-sensitive pattern, as shown in Figure 23-3. This
looping back and forth of the conductor allows for great sensitivity without increasing
the physical space the sensor occupies. Here the rest length would be 18 times longer
than the physical length of the sensor.
Figure 23-3. Typical strain-sensitive pattern
Center of gravity
The board has four legs, each of which houses a load sensor. The board uses strain gauges
similar to those discussed earlier. These gauges elongate when a force is applied to them.
The elongation changes the electrical resistance of the circuit of which the strips are a
part, and this is reported back to the controller. Figure 23-4 shows two sensor outputs.
The first is with the user standing so that her center of gravity is over the center of the
board. The second state shows what the board’s sensors would measure after the user
has shifted her center of gravity.
Button Mashing | 445
www.it-ebooks.info
25 25
15 30
Figure 23-4. Balance board example
It is easy to intuitively recognize that the center of gravity must be over the center of the
board in State A and toward the lower-right corner in State B. However, to get the exact
location of the center of gravity in State B, we’ll have to do a little more work. First things
first: we have to define a coordinate system. This is shown in Figure 23-5.
Figure 23-5. Balance board coordinate system
446 | Chapter 23: Pressure Sensors and Load Cells
www.it-ebooks.info
This coordinate system is arbitrary. If the board isn’t a perfect square, such as with the
Wii board, then the coordinates of the load cells must be changed accordingly. Now that
we have defined the location of the center of the board and the position of the load cells,
we can use a weighted average to compute the location of the user’s center of gravity.
The weight that we give each value will depend on how much of the user’s weight is on
each of the four corners. That weight will “pull” the center of gravity toward the location
of the load cells as defined in our coordinate system. How much each load cell mathe¬
matically pulls the center of gravity will be based on the weight supported at that location.
This is most easily determined via two tables, one for the x coordinate (Table 23-1) and
one for the y coordinate (Table 23-2).
Table 23-1. x coordinate weighted average
Load cell
Weight
Arm
Weight x Arm
(1,1)
30
1
30
(-1,1)
15
-1
-15
(-1,-1)
20
-1
-20
(1,-D
35
1
35
Total:
100
30
Average:
30/100 = 0.30
Table 23-1 takes the weight in each corner and multiplies that value by the value of its
x coordinate. This is equivalent to a moment. Taking the sum of those moments (30)
and dividing by the total weight gives the average value of 0.30, or .3 units to the right
in our coordinate system. The y-axis is treated similarly.
Table 23-2. y coordinate weighted average
1 Load cell
Weight
Arm
Weight x Arm 1
(i,D
30
1
30
H,D
15
1
15
(-1,-d
20
-1
-20
(i,-D
35
-1
-35
Total:
100
-10
Average:
-10/100 = -0.10
Using a similar weighted average as shown in Table 23-2, we see that the user’s center
of gravity is -0.10, or 0.1 units behind the center. If we were using this to control an
onscreen sprite, we could define a 2D direction vector based on this information.
In addition to just determining the center of gravity, you can use this information to
make educated guesses on what else the user is doing to cause the load distributions.
After computing the center of gravity, the Wii uses what Nintendo calls a motion-
identifying condition table to guess what movements the user is making. The table cor-
Button Mashing | 447
www.it-ebooks.info
relates the ratio of the sum of the load values to the body weight of the user and the
position of the center of gravity to determine body orientation. For example, the Wii
can tell if both of the user’s feet are on the board, or if the user is accelerating part of his
leg. The table provided in the Wii patent is reproduced in Table 23-3.
Table 23-3. Motion-identifying condition table
| Motion
Ratio of load value to body weight value
Position of the center of gravity 1
Right foot riding
25 to 75%
+0.01 to+1.0
Both feet riding
More than 95%
-0.7 to +0.7
Left foot riding
25 to 75%
-1.0 to-0.01
Left thigh lifting
More than 100%
+0.01 to+1.0
Right thigh lifting
More than 100%
-1.0 to-0.01
Both feet putting down
Less than 5%
Not considered
Barometers
Continuing our exploration of new user input methods, especially in the rapidly ma¬
turing mobile device gaming market, now we’ll discuss an interesting inclusion in the
latest smartphones: a barometer. Unlike buttons and balance boards, whose pressure
sensors only indirectly handle pressure, barometers directly measure the fluid pressure
that the atmosphere exerts on the sensor.
The sensors used in mobile phones today are piezoresistive microelectromechanical
systems (MEMS) and are very accurate. As shown in Figure 23-6, the sensors consist
of a void machined into a piece of silicon. The diaphragm is then bonded to a stiff
material such as steel or glass. As we are trying to measure absolute pressure, this bond
is airtight. Using a material called monocrystalline semiconductor silicon to form the
void ensures that the entire diaphragm acts much like a piezoresistive strain gauge.
Piezoresitive material
Void at reference pressure
Glass or other stiff backing
Figure 23-6. MEMS piezoresistive pressure sensor
448 | Chapter 23: Pressure Sensors and Load Cells
www.it-ebooks.info
Now we have a situation similar to the steel ball in the ocean, only this time it is a tiny
silicon ball in the ocean of air surrounding the earth. When the sensor is moved deeper
or shallower in the atmospheric ocean, the pressure on the outside of the diaphragm
changes. This causes the pressure differential to change and a force to be exerted on the
silicon diaphragm. This force causes a deflection that changes the resistance of the pie-
zoresistive material and can therefore be measured by the sensor. This part will be taken
care of by the hardware and the encoded value sent to the operating system.
To give you an example, in the Android operating system, the API has a public method,
getAltitude(float p0, float p), to determine altitude, in meters, between the sen¬
sor pressure and the pressure at sea level. It usually reads the current atmospheric pres¬
sure, p, from the sensor by listening to sensor manager callback interface method ab
stract void onSensorChanged(SensorEvent event). Here the class event holds the
sensor values, the accuracy of those values, a reference to the sensor itself, and a time-
stamp for when the event occurred. The pressure is reported in hectoPascals (hPa) or
100 N/m 2 . The sea-level pressure, p0, that this is compared to is either obtained from
an online database or is set at the constant SensorManager. PRESSURE_STANDARD_ATMOS
PHERE. As pressure at sea level changes with different weather conditions, we obtain
higher accuracy by retrieving it from a nearby airport or other weather station via the
Internet. To get the change in altitude between two points, you must repeat this process
twice as follows:
float altitude_difference =
getAltitude(SensorManager.PRESSURE_STANDARD_ATMOSPHERE,
pressure_at_point2) -
getAltitude(SensorManager.PRESSURE_STANDARD_ATMOSPHERE,
pressure_at_polntl);
At first it may seem strange for your cell phone to have a barometer in it; however, the
barometer’s ability to detect the air pressure allows you to make a good guess on your
altitude. As shown in Chapter 22, in order to determine your position via GPS you have
to solve a four-dimensional set of linear equations. The time required to solve these
equations can be dramatically decreased if you know approximately where you are to
begin with. Currently, the position of which cell phone tower your phone is connected
to is used as a starting point. Using a barometer allows the device to guess its altitude
to further reduce the time to obtain a GPS fix.
While the sensor was included for a specific purpose, it can also be adapted as an input
device. For instance, the Bosch BMP 180 currently being included in devices is accurate
to plus or minus one meter. In fact, Google Maps now provides indoor directions, in¬
cluding knowing what floor you are on in airports and shopping malls. This function¬
ality could be used to aid in the location-based gaming discussed in Chapter 22 by giving
it greater resolution in the vertical dimension. It could also be used to determine if the
user is holding the phone near her feet or her head, further augmenting the orientation
Barometers | 449
www.it-ebooks.info
sensing discussed in Chapter 21. Of course, it can also be used to help weather fore¬
casting and allow you to have real-life changes in pressure affect in-game events.
450 | Chapter 23: Pressure Sensors and Load Cells
www.it-ebooks.info
CHAPTER 24
3D Display
For all the work we’ve done to make programs’ graphics more realistic, the best we can
do is project our realistic simulations onto a two-dimensional screen. Although graphics
libraries such as Microsoft DirectX and OpenGL can provide photorealistic renderings
in real time, they still lack the ability to truly immerse the user in the works you have
so carefully created. Three-dimensional display is something that the entertainment
industry has attempted to make standard for some time. In reality, almost all “three-
dimensional” display technologies are what are technically called stereoscopic displays.
These displays use the way in which your eyes perceive depth to trick your brain into
thinking it is seeing a three-dimensional image while the display remains two-
dimensional. In contrast, displays that actually involve creating a rendering in three
dimensions are called volumetric displays. We’ll cover these later as part of our effort to
discuss emerging technologies.
Binocular Vision
The trick to displaying objects so that they appear to be three-dimensional depends on
the method by which the human brain perceives the world around it. Indeed, animals
that have two eyes engage in what is called binocular vision. Because each eye is in a
slightly different position relative to the objects it is viewing, the left and right eye provide
an image that is distinct to the brain. This is called binocular disparity. There are three
possible results when the brain encounters these two different images: suppression,
fusion, or summation. Suppression is when the brain ignores one of the images, sum¬
mation is when the brain tries to perceive both images at the same time (double vision),
and finally fusion is mixing the two images to create a depth of field. The process of
binocular fusion is something our brain learns to do when we are first born.
www.it-ebooks.info
0 %
«>r m «
_TL-v
Given the way that eyes focus light, we are born seeing the world upside
down! After a few days, our brains instinctively flip the images over so
that the motion of our hands matches the motion that we observe. There
have even been tests where people wearing glasses that invert your vi¬
sion will eventually see the images right side up. When they take off the
glasses, everything looks upside down again until their brain has time
to correct the image.
Binocular fusion is also a learned behavior of the brain. The visual cortex takes the
independent visual information from each eye and fuses it into a single image. Your
brain does this as a way of organically calculating the distance to objects so that you can
efficiently interact with the three-dimensional world. The exact process by which your
brain accomplishes this is an area of active research. In fact, researchers have found that
two images need not have any geometrical disparity in order to be fused. That is, if you
take the exact same photograph of the same object and the same angle, but with different
lighting, the shadows being cast can also cause the brain to recreate the object in three
dimensions.
Parallax is the distance an object moves between the left- and right-eye images. You can
easily demonstrate it by holding your thumb six inches from your face and closing one
of your eyes. Block some of the words on this page with your thumb. Now open that
eye and close the other one. The words that were behind your thumb should now be
visible. This is because your eyes are not in the same position, so the different angles
provide slightly different pictures of the page. This distance your thumb appeared to
move is the parallax at that distance from your eye.
Fusion is a little harder to achieve, but Figure 24-1 provides an interesting example. The
two circles are set a specific distance apart and show the top of a truncated cone coming
out of the page. The top of the cone is offset compared to the bottom. This offset is in
opposite directions, mimicking how your eyes would see it if you were directly over the
cylinder.
Figure 24-1. Cone stereopair
The best way to view the stereopair shown in Figure 24-1 is to begin by looking above
this book at a far-off object. Now lower your gaze without refocusing your eyes and
stare between the two sets of circles. With some trying, your brain should be able to fuse
the images so that there are now three sets of circles. The original two will be out of
452 | Chapter 24:3D Display
www.it-ebooks.info
452
focus, and the center set should appear to be three-dimensional. You can also get the
sets to fuse by crossing your eyes; however, this is much less comfortable than using
your eyes in their distance-viewing configuration.
Given that your brain is excellent at real-time pattern recognition, it can also compare
visual information over time to get a sense of size and relative distance. This is called
movement parallax, and it causes objects that are closer to you to appear to move faster
when you move your head than objects that are farther away. For example, if you are
driving in a car, you’ll notice that the trees appear to move faster than the moon. This
is because the trees are very close in comparison to the moon. Your brain uses this
apparent speed disparity to help conclude that the moon is very far away indeed. In the
next chapter we’ll discuss how computer algorithms attempt this sort of pattern recog¬
nition.
In fact, according to Flight Simulation (edited by J. M. Rolfe and K. J. Staples; Cambridge
University Press), the process of 3D visualization depends on the following eight major
factors.
• Occlusion of one object by another
• Subtended visual angle of an object of known size
• Linear perspective (convergence of parallel edges)
• Vertical position (objects higher in the scene generally tend to be perceived as far¬
ther away)
• Haze, desaturation, and a shift to bluishness
• Change in size of textured pattern detail
• Stereopsis
• Accommodation of the eyeball (eyeball focus)
A standard 3D graphics library is capable of giving the appearance of three dimensions
on the screen, just as any good painter on a canvas. Both standard 3D libraries and
painters do their job by recreating the first six items in the preceding list. To further the
illusion, 3D display technology simulates the seventh, stereopsis. Stereopsis is impres¬
sion of depth generated by the fact that you have two eyeballs looking at slightly different
angles. In short, the graphics library renders two different images, one for each eye, that
have a parallax shift. These images are then delivered to each eye separately. The method
by which the images are segregated varies from technology to technology. We will dis¬
cuss these in a little bit.
Binocular Vision | 453
www.it-ebooks.info
The last item on the list, accommodation of the eyeball, is the process by which your
eye changes shape to focus at different distances. By correlating the shape of your eye
with the distance to the object, accommodation works as one of the pieces of information
your brain uses to determine depth. As current 3D displays still use a 2D screen, the
eyes are still focusing on the same plane regardless of the object’s perceived depth;
therefore, the eighth item in the list is not recreated. This is why most 3D displays still
do not seem completely real. Some technologies, such as holograms and volumetric
displays, allow for accommodation of the eyeball, but usually at the expense of some
other factor. We’ll touch on these beyond state-of-the-art technologies near the end of
the chapter.
Stereoscopic Basics
There are some extra considerations when it comes using today’s 3D display technolo¬
gies to recreate the images that would usually be provided to the visual cortex by bin¬
ocular vision. Normally two eyes create two images that the brain combines with a
biological depth map. The earliest stereoscopic images were generated in the 1800s from
two photographs taken from slightly different positions. The viewer would then look at
the photos through what came to be known as a stereoscope. This device was essentially
an early example of the View-Master that some of you might remember from childhood.
While the principle of showing unique images to each eye is straightforward in this case,
it doesn’t allow group viewing and requires that the user have something pressed against
his eyes. To make 3D display something that a group of people can all experience to¬
gether and in some cases even without the aid of any headgear, we must look at some
more sophisticated methods of segregating the right and left images.
The Left and Right Frustums
If you are familiar with computer graphics, the concept of the viewing frustum is not
alien to you. If you aren’t, we’ll take a second to go over it, but it might be worthwhile
to read about it in detail before you continue. The viewing frustum is the region of space
in the model world that the camera can see from its given position in that world. In a
normal 3D graphic rendering, the frustum is clipped by a near plane that represents the
screen distance. In essence, you cannot render something closer to the user than the
screen plane. If you remember things jumping out of the screen in the last 3D movie
you saw, you can probably guess that when we are using stereoscopic rendering, this no
longer holds true. A normal computer graphics frustum is shown in Figure 24-2.
454 | Chapter 24:3D Display
www.it-ebooks.info
Figure 24-2. Normal viewing frustum
When using a stereoscopic 3D display library, we no longer have a single viewing frus¬
tum. Instead we have two cameras that are horizontally displaced from the 2D camera,
as shown in Figure 24-3.
Stereoscopic Basics | 455
www.it-ebooks.info
These two cameras, offset from the monocular camera, generate a left and right frus¬
tums. As you can see, there is a location where these two frustums intersect; this is called
the convergence distance. Objects that are placed at the convergence distance will have
the same appearance to both cameras. Note that the cameras are all pointed in the same
direction; this is called the off-axis method. This requires the frustums to be asymmetric,
which most modern graphics libraries support. Now, at first glance, it might be tempting
to toe-in the two frustums so that each camera’s frustums are symmetrical, as shown in
Figure 24-4.
Figure 24-4. Toe-in method (incorrect)
This will create workable stereopairs, but along with the horizontal parallax will intro¬
duce some vertical parallax. This can cause eyestrain to the viewer and should be avoi¬
ded. Instead, the off-axis technique should be used; it is illustrated in Figure 24-5. One
of the objects is beyond the screen in the background, and one is in front of the screen.
456 | Chapter24:3D Display
www.it-ebooks.info
You can see that if you wish to move something to greater than screen depth, the object
must be shown farther to the left than if it were at screen depth for the left eye image.
For the right eye image, the object must be shown farther to the right. If you want to
show something coming out of the screen, the opposite is true. The left eye will see the
object as farther to the right than if it were at screen depth. Also note that each object
will have a slightly different angle as well. Again, the distance between the right eye
image placement and the left eye image placement is referred to as parallax. The amount
and relative orientation of parallax is the chief way your brain creates 3D images. In
fact, the most important aspect of the physics of stereoscopic display for programmers
to understand is that there is a parallax budget that they must use wisely in developing
programs that take advantage of 3D display. This budget defines the ranges of parallax
that your viewer’s brain will be able to accept comfortably. We’ll discuss this in detail at
the end of the chapter.
For now, we’ll consider that if we were to just show the right and left images on a screen
without further work, you’d end up seeing two images with both eyes and no 3D effect
would be produced. It is paramount that the image intended for the left eye is seen only
by the left eye and vice versa. These two channels, the left and right eye, must be kept
Stereoscopic Basics | 457
www.it-ebooks.info
as separate as possible. Let’s see what the current options are for achieving such sepa¬
ration.
Types of Display
As just explained, 3D display technology depends on providing two distinct images,
one to each eye. In the next sections we’ll discuss the common techniques used today.
Complementary-Color Anaglyphs
Anyone who saw a 3D movie in the 80s remembers the cheap red and blue glasses one
had to wear to get the effect. These were complementary-color anaglyphs. An anaglyph
is the technique of encoding the separate images in a single photograph or video frame
using color filters. The method calls for two horizontally shifted images to be viewed
simultaneously. The images will contain the two images tinted in opposite colors of the
scheme. While there are many color combinations that can be used, the most common
today are red and cyan. These colors are chosen because the cyan and red filters are the
most exclusive. Red and green filters were used earlier, but the green filter allows too
much red light to leak through. This can cause what is called binocular rivalry, where
your brain has a hard time figuring out which depth map to use. One way to illustrate
this is via the simple drawing of a transparent cube, as shown in Figure 24-6.
Figure 24-6. Cube demonstrating binocular rivalry
If you focus on the cube in Figure 24-6, your brain may start to flip between interpreting
the upper face as forward of the lower face, and the lower face being forward. While
this is caused by incomplete depth cues, the same uneasy feeling can be caused when
your brain receives leaks across the two channels in a stereoscopic display. As you can
imagine, this would be pretty annoying during a video game.
As the glasses don’t require any electronics to do this, it is an example of passive 3D
technology. The major drawback of this method is that the red component of the images
is muted to the viewer. There are many improvements that can be made to the system
458 | Chapter24:3D Display
www.it-ebooks.info
to correct the color and account for some fuzziness. One example is the patented Col-
orCode 3D, which uses amber and blue filters. The advantage of this system is a nearly
full color space and a fairly good image when not viewed with the glasses.
Anaglyphs fell out of favor with movie and game producers when polarization techni¬
ques came into maturity. These produce better color reproduction and reduce eyestrain.
However, given the relatively inexpensive glasses required and that nothing special is
required of the display other than that it be capable of displaying colors, anaglyphs have
had a resurgence in printed material and online.
Linear and Circular Polarization
As polarized light plays a very important part in the largest 3D displays, movie screens,
we’ll review what polarization of light means and how to accomplish it. Light can be
thought of as an electromagnetic wave traveling through space. Let’s begin our discus¬
sion by considering a common light bulb. It emits electromagnetic waves in all directions
and is nominally “white.” An electromagnetic wave oscillates perpendicularly to its line
of travel. This is called a transverse wave. In comparison, sound waves oscillate in the
same direction they travel, creating regions of higher density. These are called longitu¬
dinal waves. Only transverse waves can be polarized because only transverse waves have
oscillation in multiple orientations. Going back to our light bulb, it is emitting “dirty”
light in that the electromagnetic waves are all at random orientations.
Most sources of electromagnetic radiation (i.e., light) are composed of many molecules
that all have different orientations when they emit light, so the light is unpolarized. If
that light passes through a polarization filter, it leaves the filter with the oscillations in
only one direction. In fact, there are two types of filters. Linear filters produce oscillations
in one direction. Circular filters (a special case of elliptical filters) create circularly po¬
larized light that rotates in an either righthand or lefthand direction. As circular filters
depend on linear filters first, we’ll discuss those now.
The simplest and most common linear filter is the wire-grid polarizer. Imagine many
very fine wires running parallel to one another with small gaps between them, as shown
in Figure 24-7. When unpolarized light hits the wires, the oscillations that are parallel
to the waves excite the electrons in the wire and move them along the length of the wire.
This phenomenon causes that component of the oscillations to be reflected. However,
the electrons cannot easily move perpendicular to the length of the wires, so the reflec¬
tion phenomenon doesn’t occur. What we are left with on the far side of the filter is a
beam of light with the oscillations all in the same direction.
Types of Display | 459
www.it-ebooks.info
Figure 24-7. Wire-grid polarization
Early 3D display systems used linearly polarized light to separate the right eye channel
from the left eye channel. However, there is one problem with using linear polarizers.
It follows that ifyou place another wire-grid polarizer after the first, with its wires rotated
90 degrees, no light will pass through! In fact, if you have an old pair of sunglasses or
3D glasses and you hold the right eye lens against the left eye lens, you won’t be able to
see anything. That is because each filter is blocking out one direction of oscillations,
preventing any light from coming through. If you rotated one of the lenses, then the
combined lens will lighten as you align the polarization directions. The problem with
these types of lenses is that if you were watching a movie and tilted your head to one
side, the same effect would occur and the image would be greatly dimmed. This means
your date could no longer rest his or her head on your shoulder while watching the
movie. Something had to be done.
Circular polarization is another form of filtering out certain orientations so that you
can control which light beams pass through which lens. However, in this case the di¬
rection of oscillation is not a single orientation but more accurately a pattern of oscil¬
lations parameterized by time. The first step to achieve circular polarization is to send
the light through a linear polarizer as just discussed. That light is then sent through what
is known as a quarter-wave filter. A typical arrangement is shown in Figure 24-8.
460 | Chapter 24:3D Display
www.it-ebooks.info
Figure 24-8. Circular polarization filter (public domain image by Dave3457; http://
commons.wikimedia.org)
First a linear polarizer rotated to 45 degrees accepts incoming light and polarizes as we
discussed earlier. The circular polarization effect is accomplished when a light wave
polarized at 45 degrees hits the filter that accepts both 0- and 90-degree oscillations. As
previously noted, this is called a quarter-wave filter. The resulting combination of 0-
and 90-degree components of the intermediate 45-degree beam results in oscillations
that turn right or left in a regular pattern. Patterns that turn counterclockwise are called
left-handed. Patterns that turn clockwise are called right-handed.
The main benefit is that the lenses create the same pattern regardless of their rotation
about the center of the lens. In other words, if you rotated the assembly shown in
Figure 24-8, meaning both lenses about the center axis, there will be no change in the
polarization. This reduces the effect of head position on the viewer’s ability to fuse the
right and left eye channels, reducing eyestrain and increasing comfort. As a side note,
it is also required for use in digital cameras, as linear polarization would affect the
autofocus and light-metering features of SLRs.
Like anaglyphs, polarized 3D systems also use glasses to separate two channels that are
projected at the same time. The first systems used two projectors, each with a different
linear polarization filter projecting on to the same screen with precise timing. As the
glasses would ahow only the correctly polarized light to be seen by either eye, the viewer
perceived binocular disparity. However, the precise timing between the projectors
would be subject to errors that cause eyestrain and binocular rivalry. Newer systems,
including RealD, use an active polarization filter fitted to the projector. However, this
is still classified as a passive system, because the glasses the user has are just normal
passive filters. In this system, there is a single filter that can change its polarization up
to 200 times a second. Every other frame is separately polarized, and binocular disparity
is experienced without the complexity of an additional projector. Although this system
Types of Display | 461
www.it-ebooks.info
uses an active filter, the glasses don’t have to actively change to separate the two channels,
so this is another example of passive technology.
The main benefit of polarized systems over anaglyphs is that they provide full-color
viewing and avoid binocular rivalry. This can increase the viewers’ comfort when they
are watching feature-length films. The disadvantages are cost and dimness. The glasses
cost much more, and the complexity of projection is increased. It is impossible to create
the effect in static media like print or web pages using standard displays. Also, because
the lens on the projector is blocking out the portions of the light that don’t have the
correct polarization, the images appear dimmer to the viewer. This can cause up to 30%
reduction in brightness and is the main point of contention for many directors.
Liquid-Crystal Plasma
The other display technologies discussed were passive technologies. The projection
carries the two channels and the glasses separate the channels, one for each eye, without
active participation from the glasses. Active technologies require that the glasses do the
work of separating the channels while the display is less important. As the gaming in¬
dustry is more sensitive to adapting 3D display technologies to work with existing
computer monitors or TVs, it has generally focused on active technologies. The most
common active technologies are based on liquid-crystal shutter glasses, or LC glasses.
The LC glasses work by exploiting a property of liquid crystals that causes them to turn
black when a voltage is applied to them. This is the same technology that creates the
eight-segment digits on a simple calculator.
Basically, every other frame being displayed is shown only to one eye, as the LC glasses
cause the lens to darken when the opposing eye’s frame is being displayed. To make sure
the glasses are preventing the correct image from being seen by the corresponding eye,
the computer broadcasts a timing signal to the glasses either over a wire or wirelessly.
At the appropriate time, the right eye lens has a voltage applied to it and the entire lens
quickly turns black. As light can enter only the left eye, that eye sees the image on the
screen. As the video being displayed moves to the next frame, this time for the right eye,
the glasses simultaneously are triggered to remove the voltage from the right lens and
apply it to the left lens. With the left lens now darkened, only the right eye sees the right
eye image. As long as this is happening very quickly,—on the order of 60 times per
second per eye, or a total refresh rate of 120 Hz—your brain can’t detect that only one
eye is seeing the information on the display at a time. Instead, it interprets it as each eye
seeing distinct images continuously, and as long as the image follows the rules we dis¬
cussed earlier, it interprets it as having depth.
462 | Chapter 24:3D Display
www.it-ebooks.info
As you can imagine, the main disadvantage of this technology for gaming would be that
you have to ensure the frame rate stays relatively high. You are now rendering twice as
many images as you normally would require. We’ll discuss more about the rendering
pipeline later. Also related to frame rate is the refresh rate of the display. As each eye is
really seeing only half the frames, the overall frame rate is half whatever the refresh rate
is on the screen. Older displays have refresh rates at 60 Hz and effectively halving that
can create issues with eyestrain; however, new displays support 120 Hz refresh rates so
that halving it still allows for smooth display. Also, the display will seem much dimmer
with the glasses on, as your eyes are seeing, on average, only half the light they normally
would. As you can see, dimness is a common problem among 3D display technologies.
The main advantage is that you don’t need a special display. As long as your display is
capable of the required refresh rate, then you can retrofit it with a pair of LC shutter
glasses and get 3D display out of it. Nvidia released such a kit in 2008, called 3D Vision,
that is relatively popular. Graphics cards are capable of automatically converting the
depth of the object in the model world into a parallax so that older games can also be
rendered in stereoscopy. This is something to consider as you design your next game
and something we’ll touch on again in a moment.
Autostereoscopy
While the newer 3D technologies are a far cry from the 3D of several decades ago, they
still rely on people wearing glasses to view the image. This means the displays can never
be used in a casual setting such as an arcade or street advertisement. Also, the younger
segment of the gaming population might break or lose the glasses. Having to put on
glasses to view the 3D images also provides a signal to your brain that what you are
about to view is optical illusion. Put on the glasses, and your brain is already thinking
that this isn’t really in 3D.
Autosterescopy endeavors to create the illusion of depth without the aid of any glasses
or other wearable device. The first and most common way it does this is by introducing
a parallax barrier between the display and the user. As discussed earlier, parallax—and
by extension, binocular disparity—is what gives our brain a main source of information
on depth. A parallax barrier uses the fact that each of your eyes sees things from a slightly
different angle to separate the two channels required for stereoscopy. Physically, the
barrier is a layer placed in or on the screen with a series of very precisely cut slits. Because
your eyes are not in the same spot, the slits reveal different pixels on the screen to each
eye. A basic illustration of this method is shown in Figure 24-9.
Types of Display | 463
www.it-ebooks.info
As shown in this figure, older screens used the slits to bar certain pixels from being seen
by placing them above the screen. Newer screens like the one on the Nintendo DS place
the barrier lower than the pixels, but before the backlight. This prevents your eyes from
receiving the light from those pixels that are being shaded by the solid spaces between
the slits. This results in a clearer image and a wider viewing angle.
Speaking of viewing angle, if the method works because your eyes aren’t in the same
spot, it is obvious that if you move your whole head then you’ll also be seeing a different
set of pixels. This is the drawback: that there is a relatively small area called the “sweet
spot” that the user must position his head relative to the screen to perceive the 3D effect.
It makes it inappropriate for movies, as only a portion of the seats would be in the sweet
spot, but it is in use for handhelds where only one user will be viewing the screen at any
given time. Another drawback is that because the slits are eliminating half the pixels
from each eye, the system reduces the effective pixel count by one-half. This causes a
reduction in resolution that can be countered by even higher pixel density.
Another method very similar to the parallax method is replacing the layer of slits with
a series of lenses that direct light from certain pixels to a certain eye. These are called
lenticular lenses and are illustrated in Figure 24-10.
464 | Chapter24:3D Display
www.it-ebooks.info
Here microscopic domed lenses are placed between the viewer and the screen. The lens
focuses the light such that only certain pixels are seen by each eye, due to the slightly
different angles by which the eyes view the lens. Benefits over the parallax barrier are
that the position of the user is less restricted and the image is brighter. With both parallax
barriers and lenticular lens arrays, it is possible to retrofit current screens with remov¬
able slide-in-place filters that allow for viewing of 3D content designed for use with
those filters. At the time of writing, several large TV manufacturers are doing active
research into widening the field of view of these technologies for use in a home enter¬
tainment environment.
Advanced Technologies
The displays we have discussed so far have all lacked some level of realism. For one, the
eye doesn’t have to refocus to observe objects at different depths, so your brain isn’t
totally fooled. Additionally, when you move around the object being projected and it
doesn’t change view, you still see the object at whatever angle the object was recorded.
If you were viewing the world through a window, you could walk to the right and see
more of the left-handed view. However, try as you might, you can’t see around the corner
of a building in a video game by moving your head at an angle of the screen. While it
might be possible to recreate that effect with some sort of head tracking, there are some
beyond-state-of-the-art technologies that could take this steps further.
One technology that is commonly thought of being able to produce 3D images is the
holograph, a staple of science fiction. It seems like we should be able to just whip up
Types of Display | 465
www.it-ebooks.info
some dynamic holographs and be done with screens all together. Who wouldn’t like to
play a sports game as if it were a table-top miniature? However, due to the way they are
recorded, holographs as we know them are static images. Once recorded, holographic
images cannot be changed. Due to their ability to encode multiple angles of viewing,
they would make wonderful display technology, and research is under way to find a
material that can hold holographic data and be rewritten fast enough to induce the
illusion of motion to the viewer. From time to time, you may see in the news some event
incorporating computer-generated imagery displayed in what is called holographic dis¬
play. These are not true holographs, but usually just a projection on a semitransparent
screen. The images are still completely flat, and the illusion of 3D comes solely from the
brain not registering the presence of the screen.
Another technology being actively researched is called integral photography. This is a
lot like the use of a lenticular lens; however, instead of linear cylinders in an array, the
lens field is more like a fly’s eye. Each lens in the array captures a complete picture from
a slightly different angle. Now, when projected through a similar integral lens, the light
forms a 4D field that the viewer sees as a 3D scene appropriate for his or her viewing
angle. If the view moves to the side, then he or she will see a new portion of the object
that wasn’t visible previously. This type of movement parallax creates very realistic 3D
experiences. The advanced displays so accurately recreate the light that recorded the
images that the eye can focus on different parts of the object (this is called the wave
front) and therefore experience accommodation of the eyeball. Recall that this is the
eighth item in our list from earlier in the chapter, and it is something the other displays
are lacking. Some crude demonstrations of this technology have been presented, and it
will be exciting to see how the research progresses.
Beyond any other method, the last one we’ll talk about takes the bull by the horns. If
you want a 3D image, just make the image three-dimensional. The other displays we
discussed all attempt to recreate 3D screens using projection from a 2D surface. There
is a group of display technologies known as volumetric that dispense with any 2D ele¬
ments and attempt to create a light field with well-defined x, y, and z coordinates. These
displays are far enough away from consumers that the definition of a volumetric display
is still being argued. One of the biggest problems with the technology will be occlusion
—that is, when an object passes in front of another object, you can’t see the object behind
the object that is closer to you. Pretty basic depth information, right? Well, if you are
attempting to create a 3D light field, it is difficult to get the light to be blocked out when
another rendered object passes in front of the original object. Simply not creating light
there won’t work, as each viewer would expect the farther object to be blocked at different
angles. There are some existing demonstrators that use lasers to excite electrons in the
air. When the lasers are focused on the same three-dimensional point, the combined
energy creates a small pocket of plasma that gives off light. These small volumes of light
are often referred to as voxels and correspond to pixels in 2D display technology. The
current resolution and refresh rate is not going to be wowing any gamers in the near
466 | Chapter 24:3D Display
www.it-ebooks.info
future, but I for one can’t wait for the day I can watch the New Orleans Saints play as a
three-dimensional table-top game.
Programming Considerations
Now that you have a background in how the current 3D display technologies work,
there are some aspects of each that you as the programmer should consider when writing
games. There are two ways to add 3D stereoscopic content to games: active stereoization
and passive stereoization. These are not to be confused with the passive and active
technologies for viewing the 3D images. The stereoization process is the method by
which the 3D images get created in the first place.
Active stereoization is the process by which the programmer creates two cameras, ren¬
dering separate images for each eye. Passive stereoization removes the requirements for
two cameras, and adds the stereoization at the GPU level. Either method is going to cost
something in performance. The worst-case cost is twice a monocular scene; however,
some elements of the scene, like the shadow map, will not require recalculation for each
eye.
Active Stereoization
Active stereoization is conceptually simpler and offers greater control over the process
of stereoization. The most naive implementation is to simply have two cameras that
render complete scenes and then pass the buffers labeled one for each eye. The buffers
are then swapped in and out with the traditional definition of frame rate being half the
actual frame rate. However, this simple implementation duplicates some elements of
the scenes that are not eye dependent.
The advantage of this technique is precise control of what each eye is seeing. This allows
the programmer to determine eye separation for each frame and could be used to ac¬
tually disorient the user as a game element. Consider a flash-bang grenade going off:
the programmer could alter the position of the cameras such that it would disorient the
user in 3D for a short period after detonation. However, this technique would cause
very real discomfort to the user, so it should be not used frequently!
The disadvantages are that the programmer is now responsible for managing an extra
camera that must be rendered for each frame. For commercial titles, this method can
be difficult considering that most games already have to be careful about how many
times they invoke the render pipeline in order to maintain playable frame rates. Having
to manage two cameras creates additional runtime burden on the program and makes
the use of existing game engines a little more difficult.
Also due to the fact that not everyone’s eyes have the same separation (intraocular dis¬
tance) and not everyone’s brain is willing to accept fabricated binocular disparity, the
program must also provide options for the user to adjust the depth and complexity of
Programming Considerations | 467
www.it-ebooks.info
the stereo effect. If you do not provide a way for the user to tune the experience, a vocal
minority will claim your game gives them a splitting headache. However, as stated be¬
fore, if you are a curious amateur, the process of moving the camera and rendering
stereoscopic images gives you a lot of insight into how the process works.
As discussed previously, we must take into account the intraocular distance of the viewer.
This is the amount of parallax we want to give objects at infinity. This distance usually
ranges between 3 cm and 6.5 cm. The large differences can arise when considering that
you must consider both adults and children when creating your 3D scene. Now it is
useful to develop a normalized measure of intraocular distance. Nvidia calls this real
eye separation and gives the following formula for the value:
Real eye separation = intraocular distance / real screen width
Note that this value will change depending on the width of the user’s screen. This value
is important because it is used as a reference for the maximum camera separation when
rendering your stereoscopic images. Separations higher than this value will cause dis¬
comfort in the viewer. In fact, for computer screens where the user is relatively close to
the screen, most people don’t feel comfortable when the camera separation is more than
half the real eye separation. This is why it is a good practice to allow the user to change
the separation as a parameter of the program. It is also important to remember that the
separation of the cameras is also the parallax value at an infinite distance. When two
images display parallax equal to separation, it is as if your eyes are perfectly parallel.
This is something that happens only when you are looking at something very far away,
like a mountaintop. If you increase parallax beyond separation, you would essentially
be asking your viewer to diverge her eyes beyond parallel. This is sort of the opposite
of crossing your eyes, and obviously is going to cause some discomfort.
As we now have an upper bound to our parallax, we can begin to create a parallax
budget. Recall that the distance at which the two frustums intersect is called convergence.
Well, at 100 times this distance, the value of parallax is 99% of separation. That means
that for objects in the scene being rendered at that depth value, they will appear flat and
all similarly far away. This is analogous to not being able to tell which peak of a faraway
mountain range is the closest.
Objects between 10 and 100 times the convergence distance have parallax that varies by
about 10%. This results in a subtle but perceptible depth differentiation. As you ap¬
proach convergence, the parallax exponentially decreases. At the convergence distance,
the parallax is equal to 0. At a distance closer to the viewer than the screen, the parallax
is negative. A distance out of the screen half that of the convergence distance creates a
negative parallax that is equal to the separation. If the object is any closer to the viewer,
her eyes verge on crossing and eyestrain becomes a problem. Now we can draw what
our budget looks like (see Figure 24-11).
468 | Chapter 24:3D Display
www.it-ebooks.info
Figure 24-11. Parallax budget
The budget scales with convergence distance and separation. You should make sure as
much of the important 3D action occurs between convergence and 10 times conver¬
gence. Your entire scene should be contained with negative convergence/2 to positive
100 times convergence.
In general, you must be most careful when trying to execute an out-of-screen effect.
These effects are very impressive to the viewer but cause the most eyestrain due to the
rapid change in parallax. Having the object first appear farther away than the screen and
then moving it closer to the user provides a context for the brain and encourages fusion.
If an object is going to be closer to the user than the screen, it is also important to prevent
that image from being clipped by the edge of the screen. That would make a portion of
the 3D object disappear, and the clipping always occurs at the convergence point. This
will give conflicting cues to the viewer and cause the 3D effect to be diminished. Given
the amount of control you need in order to prevent out-of-screen effects from causing
viewer discomfort, it is often best to use in non-player-controlled scenes.
Another difficult part of the game to render is the 2D elements. The user interface or
other menu items that have no depth are normally rendered at convergence depth.
However, there are some elements that are 2D but should be rendered at some nonzero
depth. The most important of these are mouse pointers and crosshairs. These should
be rendered at the depth of the object below them. This change in depth of the user-
controlled pointer helps maintain the idea that the objects are at different depths.
Passive Stereoization
Passive stereoization takes the responsibility for managing the stereoization process out
of the programmer s hands. The programmer sends the render pipeline a single render
command as usual, and the GPU handles generating the stereo images. Most systems
rely on heuristic subroutines in the driver to take the monocular scene and generate
binocular images. A heuristic subroutine is one that attempts to give a computer “com-
Programming Considerations | 469
www.it-ebooks.info
mon sense” about what it is trying to do to avoid having to do an exhaustive search for
solutions to an existing problem. They are algorithms not based on rigid mathematical
formulas but more like neural networks; they must be trained to do what you want them
to. These algorithms decide which elements of the scene are eye dependent and which
are not in a process that occurs entirely in the render pipeline.
It is possible for the programmer to defeat the “common sense” rules the computer is
using in the pipeline so the method is not entirely fire-and-forget but does reduce a lot
of the workload for development. One of the biggest benefits for larger game studios is
that it avoids anyone having to reprogram existing game engines. By the same token, it
allows existing games to be easily played in stereoscopic 3D. All of the preceding rec¬
ommendations apply to passive stereoization except that there should be no user-
adjustable settings in the game. Passive stereoization relies on third-party profiles that
help the GPU do the work. The user will have set up a profile with whatever stereoization
software he or she is using, such as NVIDIA’s 3D Vision. Other recommendations may
be specific to the stereoizer, and manufacturers usually publish a best practices guide.
The NVIDIA one is very helpful, and we recommend you read it if you are interested
in using stereoization in your games.
470 | Chapter24:3D Display
www.it-ebooks.info
CHAPTER 25
Optical Tracking
In previous chapters we discussed how accelerometers have changed the way that people
interact with video games. The same sort of innovation is occurring with optical sensors.
Cameras, both in visual and infrared spectrums, are being used to generate input for
games. This chapter will focus on the Microsoft Kinect for Windows SDK and give an
overview of how to make a simple game that combines optical tracking with physics.
First we’ll give a short introduction on the technologies these systems use to turn a
camera into a tracking device.
Without getting too detailed, we should start by discussing a few things about digital
cameras. First, most of us are familiar with the “megapixel” metric used to describe
digital cameras. This number is a measure of how many pixels of information the camera
records in a single frame. It is equal to the height of the frame in pixels multiplied by
the width of the frame in pixels. A pixel, or picture element, contains information on
intensity, color, and the location of the pixel relative to some origin. The amount of
information depends on the bits per pixel and corresponds to the amount of color
variation a particular pixel can display. Perhaps you’ve seen your graphics set to 16-bit
or 24-bit modes. This describes how many colors a particular pixel can display. A 24-
bit pixel can be one of 16.8 million different colors at any instant. It is commonly held
that the human eye can differentiate among about 10 million colors; 24-bit color is called
“true color,” as it can display more colors than your eye can recognize. You might also
see 32-bit color modes; these include an extra 8 bits for a transparency channel that tells
the computer what to do if this image were put on top of another image. This is some¬
times referred to as opacity or alpha.
Optical tracking and computer vision, in general, work by detecting patterns in this
wealth of pixel data. Pattern recognition is a mature field of computer science research.
The human brain is an excellent pattern recognizer. For instance, look at Figure 25-1.
Most of us can’t help but see a face in what is in reality a collection of three random
471
www.it-ebooks.info
shapes. Our brains are so primed to recognize the basic pattern of a human face that we
can do it even when we don’t want to!
o o
7 \
Figure 25-1. Four unrelated geometric entities
Computers, on the other hand, have a harder time looking at two circles and a few lines
and saying, “Hey, this is a smiling face.”
Sensors and SDKs
The modern interest in computer vision as a consumer input for computer games has
led to the development of several SDKs for performing computer-vision pattern rec¬
ognition. One such system is Kinect for Windows. Although Microsoft provides a very
high-level API with the Kinect, the downside is that you are locked into its hardware.
The popular open source alternative is OpenCV, a library of computer-vision algo¬
rithms. Its advantage is that it can use a wide variety of camera hardware and not just
the Kinect sensor.
Kinect
The Kinect was originally developed for the Xbox 360 but has recently been rebranded
to include Kinect for Windows. As console game design has high entrance requirements,
the Kinect for Windows allows more casual developers to try their hand at creating
games with optical input. The system has a hardware component, called the Kinect
sensor, and the previously mentioned Kinect SDK that does a lot of the heavy lifting for
us in terms of pattern recognition and depth sensing. The hardware component consists
of an infrared projector, infrared camera, visible light camera, and an array of micro¬
phones. The two cameras and the projector form the basis of the optical tracking system.
The projector sends out infrared light that is invisible to humans. This light bounces off
objects and is reflected back to the Kinect. The infrared camera records the reflected
light pattern, and based on how it has been distorted, calculates how far the object is
from the sensor. This exact method is carried out in the hardware of the sensor and is
proprietary. However, the patent applications reveal that a special lens projects circles
that, when reflected, become ellipses of varying shapes. The shape of the ellipse depends
on the depth of the object reflecting it. In general, this is a much-improved version of
472 | Chapter 25: Optical Tracking
www.it-ebooks.info
depth from focus, in which the computer assumes that blurry objects are farther away
than objects in focus.
As for object detection, the Kinect comes with a great set of algorithms for skeleton
direction. It can also be trained to detect other objects, but skeleton detection is really
its forte. The skeleton detection is good because of the massive amount of training
Microsoft used for the algorithms when creating the SDK. If you were to use an average
computer to run the Kinect skeleton training program, it would take about three years.
Luckily, Microsoft had 1,000 computers lying around, so it takes them only a day to run
the training simulation. This gives you an idea of the amount of training you need to
provide for consumer-level tracking in your own algorithms. The Kinect can track up
to six people with two of them being in “active” mode. For these 2 people, 20 individual
joints are tracked. The sensor can also track people while standing or sitting.
OpenCV
The OpenCV method for 3D reconstruction is, well, more open! The library is designed
to work with any common webcam or other camera you can get connected to your
computer. OpenCV works well with stereoscopic cameras and is also capable of at¬
tempting to map depth with a single camera. However, those results would not be ac¬
curate enough to control a game, so we suggest you stick with two cameras if you’re
trying to use regular webcams.
Indeed, finding depth is relatively straightforward using OpenCV. The built-in function
ReprojectImageTo3D calculates a vector for each pixel (x,y) based on a disparity map.
A disparity map is a data set that describes how pixels have changed from one image to
the next; if you have stereoscopic cameras, this essentially is the reverse of the technique
we use in Chapter 24 when dealing with 3D displays. To create a disparity map, OpenCV
provides the handy function FrndStereoCorrespondenceGC(). This function takes a
set of images, assumes them to be from a sterescopic source, and generates a disparity
map by systematically c 
Live Music Archive
Librivox Free Audio
Metropolitan Museum
Cleveland Museum of Art
Internet Arcade
Console Living Room
Books to Borrow
Open Library
TV News
Understanding 9/11