game-ai

Core Philosophy

Game Artificial Intelligence (AI) is not academic machine learning trying to predict the future or beat human chess grandmasters. Game AI is an entertainment and pacing illusion. An NPC that plays with superhuman accuracy, pixel-perfect aim, and zero hesitation is frustrating and un-fun to play against. Game AI must provide clear telegraphing, readable tactical decision-making, believable sensory perception, and dynamic difficulty calibration that gives players the thrilling sensation of outsmarting a competent opponent.

4-Step Game AI Architecture

Step 1: Decision-Making Architectures: Behavior Trees vs FSMs

1. Behavior Trees (The Industry Standard for Combat AI):
   • Built on hierarchical nodes evaluated from left-to-right:
     • Composites:
       • Sequence ($ o$): Executes children sequentially; fails if any child fails (e.g. Has Weapon $ o$ Reload $ o$ Aim).
       • Selector ($?$): Executes children until one succeeds (e.g. Flee If Low Health $?$ Melee Attack $?$ Seek Cover).
     • Decorators: Conditional gates (e.g. Invert, Time Limit, Cooldown).
     • Leaves / Tasks: Concrete executable actions (e.g. MoveToPosition, PlayAnimation).
2. Finite State Machines (FSMs):
   • Best for simple ambient creatures or discrete bosses (States: Patrol, Investigate, Chase, Attack, Flee).