分层增强学习在足球机器人比赛中的应用  

Application of Hierarchical Reinforcement Learning in Robotic Soccer

作　　者：李红娜[1] 姚分喜[1] 黄鸿[1] 

机构地区：[1]北京理工大学自动控制系,北京100081

出　　处：《计算机仿真》2005年第6期145-147,共3页Computer Simulation

摘　　要：足球机器人的研究是一项挑战性的研究领域,为了设计出智能型的球员必须涉及到计算机、人工智能、视觉及机械学等方面的研究。球员的学习能力是体现其智能的主要标志。如何在不断改变的外界环境中选取合适的动作技巧是在机器人足球比赛中的一个关键问题。该文介绍了马尔可夫决策过程,在半马尔可夫决策模型下,利用分层增强学习算法对不同层次的动作学习和选取同时进行学习。在仿真平台上进行实验,结果表明该学习方法是非常有效的。Robotic soccer is a challenging research domain because many different research areas have to be addressed in order to create a successful team of players such as computer , artificial intelligent , vision and mechanism etc. Player's intelligence can mainly be embodied in learning capacity. One problem in robotic soccer is to adapt skills and the overall behavior in a changing environment. We applied hierarchical reinforcement learning in a SMDP framework learning on all levels simultaneously. As our experiments show, learning simultaneously on the skill level and on the skill selection level is advantageous.

关 键 词：增强学习 半马尔可夫决策过程 足球机器人 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]