RECAP CSE 397/497 Topics on AI and Computer Game Programming

1
RECAPCSE 397/497 Topics on AI and Computer Game
Programming

• Prerequisites
• CSE 327 or CSE 340 or instructor consent
• Instructor consent will be immediately granted
  for
• CSE Graduate Students
• CS/CSE/CSB undergraduates that will be senior on
  the Fall/2005
• All other cases will be granted depending on the
  particular case

Héctor Muñoz-Avila
2
Course Goal
3
Path-Finding
A minimize f(n) g(n) h(n)
(Moll) Search Representations

• Grid
• Graphs
• Meshes

D dynamic A

• (Hoang)
• Path look-up matrix
• Indexed path look-up matrix
• Area-based look-up table

Precomputed Pathfinding
4
Navigation

• (Mansley)
• Qualitative Spatial Analysis
• Spatial databases
• Terrain analysis

• (Lindner)
• Navigation data can be used beyond navigation
• Low level navigation
• Jumping, climbing
• Hunting the player

5
Real-Time Strategy games

• (Schmid)
• Multi-tier AI
• Maps and AI Control
• Wall Building

• (Lee-Urban)
• Random map generation
• AI transport Units

HTN planning
6
Racing Games Sports

• (Misra)
• Contention system
• Steering
• Throttle
• Braking
• Brooks subsumption architecture

• (Gundevia)
• FSM implementation of offensive and defensive
  states (mirror)
• Dead reckoning

7
Controlling The AI Opponent (1)

• (Raim)
• FSM States, Events and Actions
• Stack Based FSMs
• Polymorphic FSM
• Multi-tier FSM

• (Hogg)
• Data-driven FSM
• Goal separate program control logic from FSM
  logic
• Scripted FSM

Robocode
( Technologies ) Advanced Flight, 4,-2,
Rad, Too, 3, 4 Alphabet, 5, 1,
nil, nil, 0, 3 ( Personalities / Goals
) Caesar, Livia, 0, 1, 1, Romans, Roman, 0, 1,
1 Montezuma, Nazca, 0, 4, 0, Aztecs, Aztec, 0,-1,
1
HTN planning
8
Controlling The AI Opponent (2)

• (Hookway)
• Ideal AI Behavior
• Coordinating behavior
• Blackboard
• Deal with obstruction
• Synthetic Adversaries
• Competence
• Taskability
• Observational fidelity
• Behavior variability

• (Grabowski )
• Movement, fire coordination
• Hierarchy of plan tactics
• Finding an Available NPC
• Availability (1N)(1O)(1P)(Q8)

HTN planning

• N of enemies in covering area

• O of enemies within range

• P of enemies threatening team

9
Controlling The AI Opponent (3)

• (Grabowski )
• Goal-Oriented Action Planning
• Alternative to FSM
• Define actions
• Inter-relations are found dynamically (planning)
• Various speed-up strategies are used

• (Xu)
• HTN Plan on level of tasks not actions
• HTNs can be used to encode game strategies
• Multi-Tier AI

Wargus
AI planning
10
Controlling the AI opponent (4)

• (Warfield)
• Several advantages of using scripts
• Modularity
• Entice players
• Main drawback developing time for the scripting
  system
• Levels of scripting
• Hardcoded (typical console game)
• Full modularity (Neverwinter nights)
• Languages
• Declarative
• Imperative

11
Adaptive AI

• (Janneck)
• Player modeling
• Simple model
• Model (attribute, value),)
• Hierarchical model
• Higher-level node value combination of childrens
  values
• Abstract node combination of concrete traits
• Issues
• Model complexity-time tradeoff
• Decouple model from game

• (Creswell)
• Decision trees are a simple representation form
• Decision trees can be learned automatically (ID3)
• One of the landmarks applications of Machine
  learning to Games

12
Adaptive AI (2)

• (Ponsen)
• Reinforcement learning to find right script
• But sometimes the problem resides in the scripts
  not the ordering
• Use evolutionary computation to improve scripts

• (Chan)
• Limitations of machine learning
• Information stored
• Pattern recognition
• During development time
• Terrain analysis
• Pattern recognition as optimization
• Pattern recognition as adaptation
• Dynamic environments

• Evolve a population (each member is a candidate
  solution)

13
Hall of Fame
14
Acknowledgements

• Jarret Raim 5 programming projects
• Marc Ponsen last programming project
• All of you
• Presentations were very good