1

AI and Game Programming

CIS 487/587

Bruce R. Maxim

UM-Dearborn

2

Selected AI Techniques

• Deterministic algorithms

• Heuristic programming

• Patterns and scripts

• Finite state machines

• Production systems

• Genetic algorithms

• Neural networks

3

Deterministic Algorithms

• Each clock tickasteroid_x += asteroid_x_velocity

asteroid_y += asteroid_y_velocity

• Asteroids follow their course until– off screen– collision occurs– blown up

4

Random Motion

fly_x = rand() % 80fly_y= rand() % 80while(~done()){

fly_count = 0fly_x_velocity = -8 + rand() % 16fly_y_velocity = -8 + rand() % 16while (++fly_count < 10){

fly_x += fly_x_velocityfly_y += fly_y_velocity

}

5

Tracking and Pursuit

if (player_x > monster_x)monster_x++

if (player_x < monster_x)monster_x--

if (player_y > monster_y)monster_y++

if (player_y < monster_y)monster_y--

6

Tracking and Evasion

if (player_x < monster_x)monster_x++

if (player_x > monster_x)monster_x--

if (player_y < monster_y)monster_y++

if (player_y > monster_y)monster_y--

7

Patterns and Scripts

• Get keys out of pocket

• Put key in door

• Open door

• Get in car

• Close door

• Put key in ignition

• Turn key to start car

8

Behavioral State Systems

• Finite state machines– give them enough states to be interesting (each

represents different goals or motives)– lots of inputs (environment attributes and other

object states)• Some states may need substates or multi-part

actions (e.g. move toward opponent and attack 20% of the time)

• Personality of characters and spheres of influence may also be factored in as probabilistic input values to state entry or transitions

9

Memory and Learning

• Memory is really just a decision to record a few details– record changes to frames– recording the last few moves

• Learning is self-improvement– neural networks– genetic algorithms– numerical approaches (e.g. Samuels & checkers)– structural approaches

10

Deciding Between AI Techniques

• Use deterministic techniques for simple behavior (e.g. rocks or missiles)

• Add randomness and patterns for smart elements (e.g. birds or spaceships)

• Use finite state machines for important game characters

• For computer controlled opponents use everything (including memory and learning)