基于CMAC的Q算法在机器人足球中的应用

摘要

RoboCup is the world＇s most influential robot soccer game, it is one of the standard research in the robotics and artificial intelligence. Simulation group is an important part of the RoboCup. As the game environment of simulation group is very complex, and the use of hand-coding to achieve Agent high-level decision-making can not could all the circumstances of the game, lack of flexibility. And it need to spend a lot of time to adjusting hand coding, the result is not satisfactory. Therefore, this paper uses machine learning to implement decisions of Agent. In this paper, we use a CMAC-based Q-learning method, and it is used in the restricted area for 2VS1 training examples, the experimental results show the feasibility and effectiveness of this method.%RoboCup是全球影响力最大的机器人足球比赛，是机器人学和人工智能及其应用的标准研究问题之一。仿真组在RoboCup中是重要的一部分。由于仿真组的比赛环境非常复杂，采用手工编码实现的Agent的高层决策无法考虑到比赛的所有情况，缺乏灵活性，并且需要花大量的时间对手工编码中的参数进行调整，结果也不是很理想。所以本文采用机器学习来实现Agent的决策。本文运用了一种基于CMAC的Q学习方法，把该方法应用在禁区内进行2VSl进攻的策略学习训练实例中，实验结果表明了本方法的可行性和有效性。