[2502.13388] Reflection of Episodes: Learning to Play Game from Expert and Self Experiences

Computer Science > Artificial Intelligence arXiv:2502.13388 (cs) [Submitted on 19 Feb 2025 (v1), last revised 5 Apr 2026 (this version, v3)] Title:Reflection of Episodes: Learning to Play Game from Expert and Self Experiences Authors:Xiaojie Xu, Zongyuan Li, Chang Lu, Runnan Qi, Yanan Ni, Lumin Jiang, Xiangbei Liu, Xuebo Zhang, Yongchun Fang, Kuihua Huang, Xian Guo, Zhanghua Wu, Zhenya Li View a PDF of the paper titled Reflection of Episodes: Learning to Play Game from Expert and Self Experiences, by Xiaojie Xu and 12 other authors View PDF HTML (experimental) Abstract:StarCraft II is a complex and dynamic real-time strategy (RTS) game environment, which is very suitable for artificial intelligence and reinforcement learning research. To address the problem of Large Language Model(LLM) learning in complex environments through self-reflection, we propose a Reflection of Episodes(ROE) framework based on expert experience and self-experience. This framework first obtains key information in the game through a keyframe selection method, then makes decisions based on expert experience and self-experience. After a game is completed, it reflects on the previous experience to obtain new self-experience. Finally, in the experiment, our method beat the robot under the Very Hard difficulty in TextStarCraft II. We analyze the data of the LLM in the process of the game in detail, verified its effectiveness. Subjects: Artificial Intelligence (cs.AI) Cite as: arXiv:2502.13388 [cs.AI]   (o...