Scholarlink

Geister is a two-player imperfect information board game in which each player hides the colors of their pieces. Capturing all of the opponent’s red pieces results in an instant loss, which makes red piece inference essential for safe aggression. In this paper, we propose a rule-based Geister AI that estimates red pieces based on the opponent’s behavior and selects moves using MiniMax search with a board evaluation function that incorporates piece advantage and estimated risk. Experimental results demonstrate that our proposed method achieves high win rates against strong baseline AIs, such as the PurpleMax, by accurately estimating red pieces and maintaining advantageous board positions throughout the game.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Rule-Based Red Piece Inference and Board Evaluation in the Imperfect Information Game Geister

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Takaya Narita,
  • Yusuke Kubota,
  • Tetsuya Suzuki

摘要

Geister is a two-player imperfect information board game in which each player hides the colors of their pieces. Capturing all of the opponent’s red pieces results in an instant loss, which makes red piece inference essential for safe aggression. In this paper, we propose a rule-based Geister AI that estimates red pieces based on the opponent’s behavior and selects moves using MiniMax search with a board evaluation function that incorporates piece advantage and estimated risk. Experimental results demonstrate that our proposed method achieves high win rates against strong baseline AIs, such as the PurpleMax, by accurately estimating red pieces and maintaining advantageous board positions throughout the game.