Source metadata

  • Type: Game AI textbook
  • Author: Ian Millington
  • Published: 2006, Morgan Kaufmann / Elsevier

Key takeaways

  • Opens by separating game AI from academic AI and warning against the complexity fallacy: in games, simple visible behaviours can read as intelligent while complex hidden systems can read as stupid.
  • Uses a practical AI model with four layers: movement, decision making, strategy, and infrastructure.
  • Gives one of the strongest classic treatments of steering behaviours, including arbitration, prediction, jumping, coordinated movement, and motor control.
  • Covers pathfinding from graphs and Dijkstra through A*, world representations, hierarchical pathfinding, interruptible planning, and movement planning.
  • Surveys multiple decision architectures: decision trees, state machines, fuzzy logic, Markov systems, goal-oriented behaviour/GOAP, rule-based systems, blackboards, and scripting.

Notable claims

  • The book argues that game AI lives under hard constraints of speed, memory, tooling, and player perception rather than abstract optimality.
  • GOAP is presented as goal-oriented behaviour extended with explicit search, including an IDA* variant.
  • Fuzzy logic is valuable in games because many design decisions are better expressed as graded desirability than as hard thresholds.

Relevance

Directly informs:

Supports:

Open questions raised

  • Which of Millington’s older infrastructure advice still maps cleanly onto modern Unity workflows, and which parts now need contemporary translation?
  • How far should the wiki go in covering older but still pedagogically useful architectures such as blackboards and rule systems?

game-ai-agent-design · steering-behaviours · pathfinding-algorithms · ai-state-machine-pattern · goal-oriented-action-planning · fuzzy-logic-for-games · blackboard · overview-artificial-intelligence-in-games