The agent uses a territorial control strategy that scores candidate moves (with optional boost) via Voronoi-style flood-fill: it simulates each move, BFS-computes distance maps from the new head and the opponent, and maximizes cells closer to itself while factoring in local liberties, connected-component size, articulation/“sealing” bonuses, and heavy risk penalties for cells/path squares the opponent can reach in 1–2 steps. It adapts weights by game phase (opening/midgame/endgame) and inferred opponent behavior (aggressive/serpentine), prefers not to reverse direction, runs a shallow beam-lookahead plus short Monte Carlo rollouts for the top candidates to gauge stability, then picks the highest-scoring move with a mild bias against using boosts unless they materially help.
Case Closed SpaceMaxxer
Voronoi Territory Scoring: Phase-based Weight Adjustment Risk Assessment: Beam Search + Monte Carlo Rollouts: Point Detection:
Updates
Leave feedback in the comments!
Log in or sign up for Devpost to join the conversation.