Procedural Content Generation for Tower Defense Games:a Preliminary Experiment with Reinforcement Learning

• Item type: Symposium(1)
• 公開日: 2021-11-06
• タイトル: Procedural Content Generation for Tower Defense Games:a Preliminary Experiment with Reinforcement Learning
• タイトル / 言語: en
• タイトル: Procedural Content Generation for Tower Defense Games:a Preliminary Experiment with Reinforcement Learning
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: Procedural Content Generation
• キーワード / 主題Scheme: Other
• キーワード / 主題: Reinforcement Learning
• キーワード / 主題Scheme: Other
• キーワード / 主題: Tower Defense
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_5794
• 資源タイプ: conference paper
• 著者所属: Graduate School of Arts and Sciences, The University of Tokyo
• 著者所属: Information Technology Center, The University of Tokyo
• 著者所属(英): en
• 著者所属(英): Graduate School of Arts and Sciences, The University of Tokyo
• 著者所属(英): en
• 著者所属(英): Information Technology Center, The University of Tokyo
• 著者名: Yueming, Xu; Tetsuro, Tanaka
• 著者名(英): Yueming, Xu; Tetsuro, Tanaka
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Although procedural Content Generation via Machine Learning (PCGML) has recently enjoyed consider-able popularity, there is little research on PCGML applied to more complicated games such as Tower Defense (TD). We trained agents to play TD levels (the solver) and generate TD levels (the generator) using reinforcement learning on a TD simulator developed in-house. We conducted the experiments of solver agents with different action spaces to find the most proper one for our task. Then we tried to generate levels, but the results showed that there is still a lot of room for improvement.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Although procedural Content Generation via Machine Learning (PCGML) has recently enjoyed consider-able popularity, there is little research on PCGML applied to more complicated games such as Tower Defense (TD). We trained agents to play TD levels (the solver) and generate TD levels (the generator) using reinforcement learning on a TD simulator developed in-house. We conducted the experiments of solver agents with different action spaces to find the most proper one for our task. Then we tried to generate levels, but the results showed that there is still a lot of room for improvement.
• 書誌情報: ゲームプログラミングワークショップ2021論文集; 巻 2021, p. 93-97, 発行日 2021-11-06
• 出版者 / 言語: ja
• 出版者: 情報処理学会