Introduction to AI in computer games

• How do you think about AI?
What is AI
2Fanafzar Game Studio

• The field of AI research was founded at a conference on
the campus of Dartmouth College in the summer of 1956
• “The science and engineering of making intelligent
machines” -- John McCarthy 1956
• The study and design of intelligent agents -- Russell &
Norvig
What is AI

• The Imitation Game (1950)
• A man (A), a woman (B), and an interrogator (C) who may
be of either sex.
Turing Test

• The Imitation Game
• We now ask the question, What will happen when a
machine takes the part of A in this game?
Turing Test

• Turing conjectured that, by the year 2000, a computer
with a storage of 10^9 units could be programmed well
enough to pass the test.
• The Turing test does not directly test whether the
computer behaves intelligently
– Some human behavior is unintelligent
– Some intelligent behavior is inhuman
• Real intelligence vs. simulated intelligence
Turing Test

• The Chinese room is a thought experiment by John Searle
which first appeared in his paper "Minds, Brains, and
Programs", published in Behavioral and Brain Sciences in
1980.
Chinese room

Chinese room

• Searle argued that software could pass the Turing Test
simply by manipulating symbols of which they had no
understanding.
• Searle concludes—the Turing Test cannot prove that a
machine can think.
Chinese room

• According to Strong AI, the correct simulation really is a
mind.
• In 1931, Kurt Gödel proved that it is always possible to
create statements that a formal system (such as an AI
program) could not prove.
Strong AI

• According to Weak AI, the correct simulation is a model
of the mind.
• Can machines think?
– boats and submarines do move through the water but
we do not call that swimming.
• Stuart Russell and Peter Norvig write: "AI researchers
have devoted little attention to passing the Turing test."
Weak AI

• The way computers "think" is vastly different from the
way a human thinks. --James Martin
• AI is faster and has a larger capacity for storage and
memory than any human.
• The largest nerves in the brain can transmit impulses at
around 90 meters per second, whereas a fiber optics
connection can transmit impulses at 300 million meters
per second, more than 3 million times faster.
Alien intelligence

• “A computer program is said to learn from experience E
with respect to some class of tasks T and performance
measure P, if its performance at tasks in T, as measured
by P, improves with experience E.” --Tom M. Mitchell
• REcognition, classification
• Online and Offline learning
• Supervised, Unsupervised, Reinforcement
Machine Learning

• Classification
Machine Learning - Supervised

• Clustering
– We don’t know number of classes
Machine Learning - Unsupervised
X Y
12.37 15.64
22.8 7.8
34 17
91 50
11.9 17
44 19
80 45
21 9
33.31 16.5
79 39
… …

• Clustering
Machine Learning - Unsupervised

• The goal of a reinforcement learning agent is to collect as
much reward as possible.
• Highly related to dynamic programming techniques
• Most famous technique is Q-learning
• Reinforcement Learning in First Person Shooter Games
– IEEE TRANSACTIONS ON COMPUTATIONAL INTELLIGENCE AND AI IN GAMES,
VOL. 3, NO. 1, MARCH 2011
• High-level Reinforcement Learning in Strategy Games
– International Conference on Autonomous Agents and Multiagent Systems
Machine Learning - Reinforcement

• A simple idea: use the theory of evolution as an
algorithm.
• A Population of Individuals
• Swarm intelligence
– Ant colony optimization
– Particle swarm optimization
– Bees algorithm
– Cuckoo search
Evolutionary Computing

• Individual = Chromosomes
• Mutation, Selection, and Crossover.
• Operating on dynamic data sets is difficult
• Tendency to converge towards local optima
• Randomness
Genetic Algorithm

• Simplified models of neural processing in the brain
Neural Networks

• Multilayer Perceptrons
Neural Networks

• How to train
• Black box
• Over fitting
• Computationally expensive
• Evolving Neural Controllers using GA for Warcraft 3-Real
Time Strategy Game
– 2011 Sixth International Conference on Bio-Inspired Computing
Neural Networks

• A good book about
GA and NN in games
Neural Networks

Do we really need AI in computer games?
What we expect from game AI?

• Efficiency
• Ease of Debugging (randomness)
• We don’t need general problem solver
• Believability
– We don’t need human level intelligence
– It doesn't really matter how NPC intelligence is
achieved, as long as the creatures in the game appear
believable.(weak AI)

• Video games provide a rich test bed for artificial
intelligence methods
• Designers need to control the behavior of NPCs
– Explicit control
– Implicit control
• It is very genre specific
• Avoid artificial stupidity

• Agent cycle
Agents as NPCs
THINKSENSE ACT

• Agent cycle
Agents as NPCs
MAP
GEOMETRY
ENTITIES
. . .
THINK ACT

• Agent cycle
Agents as NPCs
REMEMBER
REASON
BEHAVE
SENSE ACT

• Agent cycle
Agents as NPCs
ANIMATE
NAVIGATE
. . .
THINKSENSE

• Rule : if (condition) then action
• Production Rule System comprised of a database of rules,
each rule consists of an arbitrarily complex conditional
statements.
• They are fairly uncommon approach.
Rule Based

• Search Methods, discovering a sequence of actions or
states within a search space that satisfy some goal
• Goal-oriented behavior is still fairly rare in games.
Goal Oriented

• Everything in game world is triangle
• In door / Out door
• Path finding still is a problem
– Some pathfinding bugs(Video)
Introduction to Path planning

• Graph Theory – shortest path
• Single Source shortest path
• All pairs shortest path – Floyd
– Store the result
• Heuristic F(n) = D(n) + H(n)
A*/Dijkstra

• Near-Optimal Hierarchical Pathfinding,
– A. Botea, M. Muller, and J. Schaeffer, Journal of Game
Development, Volume 1
Hierarchical Pathfinding

• Grid/Tile Base
– Fast
– Easy to develop
– Memory Inefficient
– 2D and strategy games
Reviewing some pathfinding methods

• Waypoint graphs
– Manual

• Waypoint graphs
– Automated : Point of visibility

• Worlds require a ridiculous number of waypoints
• Difficult dynamic obstacle avoidance, if not impossible
• Is not shortest path – not optimal
• Impossible to do path-smoothing
• Zig Zag path

Fanafzar Game Studio 40
Inefficient

• Convex Polygons
– Manual/Automated

• Shortest path - Optimal
• Smaller Search Space

OK, But how
to
implement?

• Mesh simplification
• Rendering techniques
• Flood filling with AABBs (UDK)
• Voxelization
• Check Mikko Monone’s work, RecastNavigation
– http://digestingduck.blogspot.com

• The general process is as follows:
1. Voxelization
2. Generate Regions
3. Generate Contours
4. Generate Polygon Mesh
5. Generate Detailed Mesh
Navigation mesh generation process

1. Voxelization

2. Generate Regions

Douglas-Peucker simplification

4. Generate Polygon Mesh
– Triangulation
– Merge to Convex Polygon
– Benefits of Convex Polygon

• Path find with A*
– Graph nodes are convex polygons
– Corridor map
– A* is not complex to implement
– There are many optimization techniques

• Pathfinding is Not A Star, AUTODESK® KYNAPSE ®
MIDDLEWARE WHITE PAPER
– Path smoothing
– Path following
– Deal with other NPCs
– Deal with dynamic evolutions of game world

• Velocity obstacle
• There are some variations
– RVO, NLVO, FVO, HRVO, NHRVO, PVO
• Mikko uses RVO(Reciprocal Velocity Obstacles)

• Voxelization also used in Cover selection, Jumps,
Camera movement
– Automatic annotations in Killzone 3 --Mikko Mononen,
Paris Game AI Conference 2011

Navigation mesh, Zorvan Integration

• Generation Types
– Solo
– Tiled Navmesh
• Dynamic loading
• Deal with dynamic obstacles
• Off-Mesh Connections
• Convex polygons
Navigation mesh, Zorvan Integration

• How to simulate movement of intelligent objects
like bird, animals, cars, etc. --Craig Reynols
• Stanley and Stella in: Breaking the Ice (1987)
Steering Behaviors

• Simple Vehicle Model
– Mass scalar
– Position vector
– Velocity vector
– max_force scalar
– max_speed scalar
Steering Behaviors

• Seek
• Flee
• Flocking
• Pursuit
• Arrival
• Obstacle avoidance
• Path follow
• Leader follow
• Hide
Steering Behaviors

• Seek Vs Flee
Steering Behaviors

• Seek
Vector2D SteeringBehaviors::Seek(Vector2D TargetPos)
{
Vector2D diff = TargetPos - m_pVehicle->Pos();
Vector2D DesiredVelocity =
Vec2DNormalize(diff) * m_pVehicle->MaxSpeed();
return (DesiredVelocity - m_pVehicle->Velocity());
}
Steering Behaviors

• Obstacle avoidance
– Only circles
Steering Behaviors

• Combining Steering Behaviors
– Weighted Truncated Sum
– Weighted Truncated Running Sum with Prioritization
– Prioritized Dithering
• Open Steer
Steering Behaviors

Advantages
– Simplicity
– Reliability
– Predictability
– Efficiency
Disadvantages
– Local traps
• Oscillation
– Realism
• Jagged paths
– Scalability
Steering Behaviors

• Pure scripting!
• Structured
– FSM
– HFSM
– Behavior Tree
Decision Making

• Theory (Simplified)
– A set states, S
– An input vocabulary, e
– Transition function, T(s, e)
• Map a state and an input to another state
FSM
Patrol
(idle)
Combat
Return to
post
Pursue
Player Seen
Near PlayerFar from post
Reach post

• Finite State Machine (FSM) is the Most Commonly
used Game AI Technology Today
– Simple
– Efficient
– Easily extensible
– Powerful enough to handle a wide variety of situations
• Decisions only depend on current state
FSM

• Hard Coded
– Switch Statement
– Function pointers
– Polymorphism (State Pattern)
• Interpreted
– Data Driven
– Scripted
• Compiled
– machine code
– Generating source code
FSM

• Function Pointer-Based, Embedded Finite-State Machines
– Chapter 3.1, Game programming gems 1
• A Finite-State Machine Class
– Chapter 3.3 Game Programming Gems 3
FSM

• Each state can be a complete state machine in its own
right
• Original Paper
– Statecharts: A Visual Formalism for Complex Systems D. Harel
Science of Computer Programming 8, 1987
• Clustering states (XOR)
• Concurrency (AND)
HFSM

• Example
HFSM
Patrol
(idle)
Return to
post
Pursue
Player Seen
Near PlayerFar from post
Reach post
Combat
Attack 1 Attack 2
Success

• Cluster states
• The semantic of D is exclusive-or (XOR) of A and C
HFSM
A
C
B
α
β
δ
γ(cond)
D

• Parallel (AND combination)
HFSM
A
C
α β
E
G
F
α
δ
γ μ
A DY

Iranvij AI Editor

• Halo 2 [Bungie Software, 2004] was one of the first high-
profile games for which the use of behavior trees.
• Instead of a state, the main building block of a behavior
tree is a Task
• Conditions, Actions, and Composites
Behavior tree

• Selector
Behavior tree
terminate
keep trying
?

• Sequence
Behavior tree
keep going
bail out

• Example
Behavior tree
?
Door
Open?
Move
(into room)
Move
(to door)
Open
door
Move
(into room)
Condition Action

• Refactored tree
Behavior tree
?
Door
Open?
Move
(to door)
Open
door
Move
(into room)

• Decorators (such as Invertor)
• Random Selector
• Random Sequence
• Parallel

• Procedural Animation
• Procedural level generation
• Dynamic game difficulty balancing
Next Generation AI for game

• Artificial Intelligence For Games, Second Edition, Ian
Millington, John Funge
• Programming Game AI by Example, Mat Buckland
• Game programming gems series
• AI Game Programming Wisdom series
• AI Game Development: Synthetic Creatures with Learning
and Reactive Behaviors, Alex J. Champandard
• http://aigamedev.com/
• http://www.ai-blog.net/archives/000183.html
• http://www.critterai.org/nmgen_study
References

• http://digestingduck.blogspot.com/
References

Thank You
hojjatjafary@yahoo.com

Introduction to AI in computer games

More Related Content

What's hot (20)

Viewers also liked (15)

Similar to Introduction to AI in computer games (20)

Recently uploaded (20)

Introduction to AI in computer games