摘要
全局游戏策略GGP(General Game Playing)旨在开发一种没有游戏经验支撑下能够精通各类游戏的人工智能。在原有强化学习算法研究的基础上,提出一种基于经验的简化学习方法,通过对游戏状态的筛选和游戏经验的归纳,从而降低决策对经验数量的需求,提高决策效率,并能达到指定胜利、平局或失败的游戏目标。通过在三种不同的游戏规则下与玩家进行游戏比赛实验表明,该学习方法能有效地达到预期结果。
General Game Playing aims at developing game playing agents that are able to become proficient at playing a variety of games without specific preparatory game experience.After studying existing reinforcement learning algorithms,the paper puts forward an experience generation shortcut learning method.Through game status selection and game experience conclusion,decision making requires less experience and enhances efficiency.The destined goal of the game,either victory,or draw,or defeat,can be reached.Experiments are carried out with three game rules respectively against human players.It is proven that the proposed learning method can effectively meet expectations.
出处
《计算机应用与软件》
CSCD
北大核心
2012年第1期253-256,275,共5页
Computer Applications and Software
关键词
全局游戏策略
人工智能
强化学习
General game playing Artificial intelligence Reinforcement learning