Introduction to Game AI

CS 395 Game Design

Spring 2002
 Expectations on term projects
Designing and implementing a full game from
scratch can be too hard
Exceptions: Text-based interactive fiction, arcade
games,
What to do?
 Solution One: Leverage
Were supplying toolkits and resources for some
projects
Neverworld3
Self-explanatory simulator compiler
There are lots of resources around if you look
Game engines
Graphics frameworks
Games with available source code (e.g., FreeCiv)
Building on top of such software is strongly
recommended.
 Solution Two: Extension an existing
game
Extensions to existing games can be acceptable
(aka mods)
Must involve significant design work, including
analysis of tradeoffs and portfolio of experiments
generated in the course of design
Must be substantially larger than a typical homework
problem
 Term Project To Do List
Identify your teammates early
Identify your project area early
Request for Proposals will be made very soon
 Overview
Why AI is important for games
And why it needs improving
Roles for AIs in games
The opponent
Characters
The World
In-depth look: Tactical decision making in first-
person shooters
Thanks to John Laird, Mike VanLent, for their GDC
2001 tutorial material
 Why AI is important for games
Essential to the modeled world
NPCs of all types: opponents, helpers, extras,
Unrealistic characters reduced immersion
Stupid, lame behaviors reduced fun
Superhuman behaviors reduced fun
Until recently, given short shrift by developers
Graphics ate almost all the resources
Cant start developing the AI until the modeled world
was ready to run
AI development always late in development cycle
Situation rapidly changing
AI now viewed as helpful in selling the product
Still one of the key constraints on game design
Interview with Soren Johnson, www.gamespy.com

GameSpin: What about the [civ3] AI? One of the

complaints that players have always had about the AI
is that it cheats. Does it still cheat?

Johnson: The AI has been totally reworked. We started from

scratch. We stretched out the difficulty levels. Chieftain is easier
than it was in Civ II and Deity is now harder. Does the AI cheat?
Yes, but sometimes in favor of the player! Below Prince level it
cheats for the player, and above Prince level it cheats against
the player. At Prince level there is no cheating.

Right now, no one at Firaxis can beat the AI at Deity level,

though we certainly expect players to find ways to do so within a
few months of the game being released.

When we rewrote the AI we threw out the old AI completely. We

also looked at some popular tactics like ICS -- infinite city sprawl
-- that some players used to beat the AI in Civ II at Deity level.
These won't work in Civ III.
 AI in game marketing
It may be hard to believe that the future of 21st
century art is represented by a giant bipedal tiger
who farts, break dances and flings livestock
around when he's bored.
Review of Black and White in Salon.com, 4/10/2000
The most versatile 3D Batman yet, with over 500
animated movements, special fighting moves and
a multi-functional cape with its own A.I.
from www.xbox.com
 Game flaws engendered by bad AI
Mindless hordes of opponents
Life is much, much easier thanks to random numbers
Rampant cheating by computer opponents
Focus on violence as principle mode of interaction
 Roles for AIs in games
The opponent
Characters
The world
 The opponent
Emphasis: Outsmarting you
Example: Classic board games
Questions:
What problems do they have to solve?
How do they work?
 Opponents in classic board games
Move generator figures out what moves are
possible
Static evaluator figures out how good each move
is, based on changes in board position
Search algorithms look ahead, in a simple form of
mental simulation, to figure out the best
alternative based counter-moves and counter-
moves
 Example

Current board position

 Example Your possible
moves, as produced
by move generator

6 18 -2

How good each move looks in

terms of its immediate
results, as computed by the
static evaluator
 Example
Your opponents
possible
countermoves

8 2 -100 25 18 12

What position these

countermoves leave you in
 This process
Example
Use minimax
can be to estimate
continued your best
until resources move
run out
 Minimax Example
Run static
evaluator
on leaves

8 -10 3 5 -98 210 108 22 -12 18 -40

 Minimax Example
Propagate
scores upward
by taking max
when your turn

210 18
8 5 -98 22

8 -10 3 5 -98 210 108 22 -12 18 -40

 Propagate
Minimax Example
scores upward
taking min
when opponents
turn
18
5 -98

210 18
8 5 -98 22

8 -10 3 5 -98 210 108 22 -12 18 -40

 Minimax Example
Select best
option based
on deeper
estimate

18
5 -98

210 18
8 5 -98 22

8 -10 3 5 -98 210 108 22 -12 18 -40

 Search issues
Lots of techniques for more efficient search
Alpha-beta pruning, book openings, stability
measures, conspiracy numbers
Basic problem is still exponential
Search and brute force only go so far
Q: Why did Deep Blue win?
Knowledge/search tradeoff
Simon & Chase experiments: Chess experts use spatial
memory for board positions
Standard patterns as encoding lessons from
experience/deeper search?
Opponents in turn-based strategy games
Examples: Many strategy war games, Civilization-
style games
What problems do they have to solve?
How are they similar to, and different from, board
games?
 Characters
Examples: First-person shooters, adventure/action
games
What problems do they have to solve?
 The World
Common trick: Share AI computations between
computer characters
Pathfinding and navigation
Awareness of player actions and alignment
Strategic thinking
Examples: More than you might think
 Types of Behavior to Capture
Wander randomly if dont see or hear an enemy.
When see enemy, attack
When hear an enemy, chase enemy
When die, respawn
When health is low and see an enemy, retreat

Extensions:
When see power-ups during wandering, collect them.
Execution Flow of an AI Engine

Sense

Finite-state machines
G Decision trees
A
Neural nets
M Think ?
Fuzzy logic
E
Rule-based systems
Planning systems

Act
 Conflicting Goals for
AI in Games

Goal Reactive
Driven

Human
Knowledge
Characteristics
Intensive

Low CPU & Fast & Easy

Memory Development
Usage
 Complexity

Complexity of Execution
How fast does it run as more knowledge is added?
How much memory is required as more knowledge is added?
Complexity of Specification
How hard is it to write the code?
As more knowledge is added, how much more code needs
to be added?
 Expressiveness of Specification
What can easily be written?
Propositional:
Statements about specific objects in the world no
variables
Jim is in room7, Jim has the rocket launcher, the rocket
launcher does splash damage.
Go to room8 if you are in room7 through door14.
Predicate Logic:
Allows general statement using variables
All rooms have doors
All splash damage weapons can be used around corners
All rocket launchers do splash damage
Go to a room connected to the current room.
 Example FSM

Attack Events:
E, -D
E=Enemy Seen
E S=Sound Heard
E
-E D=Die
E D
Wander Chase
-E, -S, -D S
S, -E, -D
-S
D D
-E
S
Code
Spawn
D

Action (callback) performed when a transition occurs
 Example FSM

Attack Events:
E, -D
E=Enemy Seen
E S=Sound Heard
E
-E D=Die
E D
Wander Chase
-E, -S, -D S
S, -E, -D
-S
D D
-E
S
Problem: No transition
Spawn from attack to chase
D
 Example FSM - Better

S
Events:
Attack Attack-S
E, -D, -S E, -D, S E=Enemy Seen
-S

-E
S=Sound Heard
E D E
D=Die
-E
E D
Wander Chase
-E, -S, -D S
S, -E, -D
-S
D D
-E
S
Spawn
D
 Example FSM with Retreat
S Attack-ES
Retreat-S
Attack-E E,-D,S,-L
L -E,-D,S,L
E,-D,-S,-L -S
L -L E Events:
-E
E E -L
-L
E=Enemy Seen
Wander-L Retreat-ES
-E,-D,-S,L E L E,-D,S,L S=Sound Heard
-S
-L D=Die
L S
Retreat-E L=Low Health
Wander -E -E E E,-D,-S,L
-E,-D,-S,-L
D D Each feature
D D Chase with N values
-E,-D,S,-L
Spawn can require N
D S
times as many
(-E,-S,-L)
states
 Hierarchical FSM
Expand a state into its own FSM
Attack

Wander E/-E

Pick-up
Powerup

S/-S Chase
Start
Turn Right

Go-through
Door
Die

Spawn
Non-Deterministic Hierarchical
FSM (Markov Model)

No enemy Wander

Attack

Approach
.3
Aim &
.3
Slide Right
& Shoot .3 Start
.4 .3
Aim &
.4
Slide Left
Aim &
& Shoot
Jump &
Shoot Die

Start
 Implementation
Compile into an array of state-name, event event
state-name := array[state-name, event]
state
Uses state-name to call execution logic
Add buffers to queue up events in case get
simultaneous events

Hierarchical
Create array for every FSM
Have stack of states
Classify events according to stack
Update state which is sensitive to current event
 FSM Evaluation
Advantages:
Very fast one array access
Can be compiled into compact data structure
Dynamic memory: current state
Static memory: state diagram array implementation
Can create tools so non-programmer can build behavior
Non-deterministic FSM can make behavior unpredictable
Disadvantages:
Number of states can grow very fast
Exponentially with number of events: s=2e
Number of arcs can grow even faster: a=s2
Propositional representation
Difficult to put in pick up the better powerup, attack the
closest enemy
 References
Web references:
www.gamasutra.com/features/19970601/build_brains_into_games.
htm
csr.uvic.ca/~mmania/machines/intro.htm
www.erlang/se/documentation/doc-
4.7.3/doc/design_principles/fsm.html
www.microconsultants.com/tips/fsm/fsmartcl.htm
Deloura, Game Programming Gems, Charles
River Media, 2000, Section 3.0 & 3.1, pp. 221-
248.