Playing Puyo Puyo with Deep Reinforcement Learning

カテゴリ: 部門大会

論文No: SS2-3

グループ名: 【C】2024年電気学会電子・情報・システム部門大会

発行日: 2024/08/28

タイトル(英語): Playing Puyo Puyo with Deep Reinforcement Learning

著者名: 宮國 雅士（千葉大学）,中間 公啓（千葉大学）,小圷 成一（千葉大学）

著者名(英語): Masashi Miyaguni (Chiba University),Kimihiro Nakama (Chiba University),Seiichi Koakutsu (Chiba University)

キーワード: 深層強化学習|ゲームAI|ぷよぷよぷよぷよ|Deep Reinforcement Learning|Game AI|Puyo Puyo

要約(日本語): In recent years, various video games have become competitive and are now being played as e-Sports. Puyo Puyo, a nationally popular puzzle game, has also become an e-Sport and is currently gaining momentum. On the other hand, development of game AI that applies reinforcement learning, a type of machine learning, to games is underway. AlphaGo Zero, an AI that uses deep reinforcement learning, has attracted a great deal of attention for its victory over the world champion in the highly difficult board game of Go. Although AI using deep reinforcement learning has also been studied in Puyo Puyo, it has not been able to construct large chains. In this study, we apply deep reinforcement learning to Puyo Puyo and attempt to create a game AI that can construct large chains.