AI Behavioral Systems for Sports Games

Sammanfattning

This thesis presents a hybrid artificial intelligence system for team-based sports games, developedfor the mobile title “Pocket Hockey” (Gold Town Games). The existing finite-state machinecontrolling non-playable skaters produces rigid behavior that is difficult to tune and extend.The replacement system separates tactical positioning from action selection: the Hungarianmethod assigns skaters to formation roles by minimising a cost function that combines distanceand character affinity, while utility theory scores candidate actions—pass, shoot, tackle, mark,and others—against contextual features and individual skill attributes. Post-processing bypower curve, minimum threshold, and hysteresis cooldown eliminates flicker and noise-drivenselections. The entire system is implemented in the Unity engine as a drop-in replacement forthe prior AI module. All behavior weights are exposed as tunable parameters, allowing difficultyand play-style adjustments without code changes. The result is a modular, data-drivenarchitecture that coordinates multiple agents in real time and produces context-responsive teambehavior.