基于模糊小波网络的强化学习及其在多机器人决策策略中的应用  

Reinforcement learning based on FWNN and its application in decision-making for multi-robot cooperation

作　　者：段勇[1] 李程[1] 徐心和[2] 

机构地区：[1]沈阳工业大学信息科学与工程学院,沈阳110870 [2]东北大学信息科学与工程学院,沈阳110004

出　　处：《高技术通讯》2013年第4期407-412,共6页Chinese High Technology Letters

基　　金：国家青年科学基金(60905054);辽宁省高等学校杰出青年学者成长计划(LJQ2011006)资助项目

摘　　要：给出了一种基于模糊小波神经网络(FWNN)的强化学习方法,并研究了应用该方法解决多机器人足球比赛中的决策策略问题。首先,使用FWNN来实现强化学习状态空间到动作空间的映射,从而解决大规格或连续状态空间所导致的学习速度过慢甚至难以收敛等问题。然后,研究了提出的方法在机器人足球比赛的复杂决策策略学习中的应用,证明机器人球员能够通过学习掌握根据比赛状态信息选择合理动作的能力。最后,通过实验验证了该学习方法的有效性,它能够满足机器人足球比赛的需要。A reinforcement learning （RL） algorithm based on fuzzy Furthermore, its application in selection of decision-making FWNN was used to perform the mapping from the state space to wavelet neural networks （FWNN） was proposed. strategies for robot soccer was studied. Firstly, a the action space of RL, consequently, the problems of slow learning and difficult convergence caused by the large or continuous state space were solved effectively. Then, the application of the presented method in learning of decision-making strategies for robot soccer was stud- ied, achieving the result through learning, the robot players can master the ability of selecting actions based on their states in the game. Finally, the effectiveness of the presented method was verified by experiment. The experimental result shows that it can meet the demands of robot soccer.

关 键 词：强化学习(RL) 模糊小波神经网络(FWNN) 机器人足球比赛 动作选择 决策 

分 类 号：TP242[自动化与计算机技术—检测技术与自动化装置]