基于改进的Q学习的RoboCup传球策略研究  

Research of RoboCup Pass Strategy Based on Improved Q-Learning

作　　者：周勇[1] 刘锋[1] 

机构地区：[1]安徽大学计算机科学与技术学院,安徽合肥230039

出　　处：《计算机技术与发展》2008年第4期63-66,共4页Computer Technology and Development

基　　金：国家自然科学基金(60273043);安徽省自然科学基金(050420204)

摘　　要：模拟机器人足球比赛(Robot World Cup,RoboCup)作为多Agent系统的一个理想的实验平台,已经成为人工智能的研究热点。传统的Q学习已被有效地应用于处理RoboCup中传球策略问题,但是它仅能简单地离散化连续的状态、动作空间。提出将神经网络应用于Q学习,系统只需学习部分状态-动作的Q值即可获得近似连续的Q值,就可以有效地提高泛化能力。然后将改进的Q学习应用于优化传球策略,最后在RobCup中实现测试了该算法,实验结果表明改进的Q学习在RoboCup传球策略中的应用,可以有效提高传球的成功率。As the ideal experimental platform of multi-agent system, RoboCup（Robot World Cup） has become the research center of artificial intelligence. Traditional Q-learning dispersed sequential state and action simply on resolving the problem about pass strategy in RoboCup environment. Puts forward a method that neural network is applied to Q-learning, system would output sequential state-action by learning Q-value based on partial state-action and improve generalization ability effectively. The method based on improved Q-learning is proposed to optimize the pass strategy and test the algorithm in RoboCup environment. The experiment shows that improved Q-learning can improve pass efficiency effectively in RoboCup environment.

关 键 词：ROBOCUP 神经网络 Q学习 智能体 传球策略 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]