全部 标题 作者
关键词 摘要

OALib Journal期刊
ISSN: 2333-9721
费用:99美元

查看量下载量

相关文章

更多...

基于CMAC网络Sarsa(λ)学习的RoboCup守门员策略

Keywords: RoboCup仿真组足球比赛,CMAC神经网络,泛化,Sarsa(λ)学习算法,最优策略

Full-Text   Cite this paper   Add to My Lib

Abstract:

针对RoboCup仿真组足球比赛场上状态复杂多变、同时供决策的信息大多为连续变量、智能体利用现有信息通常无法判断当前状态下最优动作的问题,以守门员为例,首先利用CMAC神经网络对连续状态空间泛化,然后在泛化后的状态上,采用Sarsa(λ)学习算法获取守门员的最优策略.通过在RoboCup仿真平台上进行仿真,实验结果表明,采用基于CMAC的Sarsa(λ)学习算法的守门员,经过一定时间的学习后,防守时间显著增长,防守效果明显优于其他算法,验证了本文所提方案的有效性.

References

[1]  段勇,杨淮清,崔宝侠,等.强化学习在足球机器人基本动作学习中的应用[J].机器人,2008,30(5):453-459.DUAN Yong,YANG Huai-qing,CUI Bao-xia,et al.Application of reinforcement learning to basic actionlearning of soccer robot[J].Robto,2008,30(5):453-459.(in Chinese)
[2]  NODA I,MATSUBARA H,HIRAKI K,et al.Soccerserver:a tool for research on multiagent systems[J].Applied Artificial Intelligence,1998,12(2/3):233-250.
[3]  WEIGEL T,GUTMANN S,DIETL M,et al.CSFreiburg:coordinating robots for successful soccer playing[J].IEEE Transactions on Robotics and Automation,2002,18(5):685-699.
[4]  李人厚.智能控制理论和方法[M].西安:西安电子科技大学出版社,1999:116-120.
[5]  SUTTON S,BRATO G.Reinforcement learning:anintroduction[M].Cambridge:A Bradford Book,1998:4-6.
[6]  STONE P,SUTTON S.Scaling reinforcement learningtoward RoboCup soccer[C]∥The EighteenthInternational Conference on Machine Learning,WilliamsCollege.Williamstown:Morgan Kaufmann,2001:537-544.
[7]  WHITESON S,TAYLOR E,STONE P.Empirical studiesin action selection with reinforcement learning[J].Adaptive Behavior,2007,15(1):33-50.
[8]  KOK J.UvA trilearn 2003-soccer simulation team[EB/OL].[2010-07-18].http:∥staff.science.uva.nl/~jellekok/robocup/2003/.

Full-Text

Contact Us

service@oalib.com

QQ:3279437679

WhatsApp +8615387084133