Hierarchical Reinforcement Learning with OMQ

摘要：

A novel method of hierarchical reinforcement learning, named OMQ, by integrating Options into MAXQ is presented. In OMQ, the MAXQ is used as basic framework to design hierarchies experientially and learn online, and the Option is used to construct hierarchies automatically. The performance of OMQ is demonstrated in taxi domain and compared with Option and MAXQ. The simulation results show that the OMQ is more practical than Option and MAXQ in partial known environment.

关键词： hierarchical reinforcement learning Option MAXQ.

作者: Jing Shen Haibo Liu Guochang Gu

作者单位: School of Computer Science and Technology, Harbin Engineering University Harbin 150001, China

会议类型: 国际会议

会议名称: Firth IEEE International Conference on Cognitive Informatics(第五届认知信息国际会议)

会议地点: 北京

会议语种:英文

页码: 584-588

在线出版日期: 2006-07-17（万方平台首次上网日期，不代表论文的发表时间）