计算机工程与应用 ›› 2007, Vol. 43 ›› Issue (10): 19-21.

• 博士论坛 • 上一篇    下一篇

基于强化学习的未知环境多机器人协作搜集

赵杰 姜健 臧希喆   

  1. 哈尔滨工业大学 机器人研究所 多传感器集成及控制研究室 哈尔滨工业大学 机器人研究所 多传感器集成及控制研究室 哈尔滨工业大学 机器人研究所 多传感器集成及控制研究室
  • 收稿日期:2006-12-25 修回日期:1900-01-01 出版日期:2007-04-01 发布日期:2007-04-01
  • 通讯作者: 姜健

Cooperative Multi-Robot Foraging Based on Reinforcement Learning in Unknown Environment

Zhao Jie Jian JIANG Zang Xizhe   

  • Received:2006-12-25 Revised:1900-01-01 Online:2007-04-01 Published:2007-04-01
  • Contact: Jian JIANG

摘要: 针对多机器人协作复杂搜集任务中学习空间大,学习速度慢的问题,提出了带共享区的双层强化学习算法。该强化学习算法不仅能够实现低层状态-动作对的学习,而且能够实现高层条件-行为对的学习。高层条件-行为对的学习避免了学习空间的组合爆炸,共享区的应用强化了机器人间协作学习的能力。仿真实验结果说明所提方法加快了学习速度,满足了未知环境下多机器人复杂搜集任务的要求。

关键词: 搜集任务, 多机器人系统, 强化学习, 协作

Abstract: To reduce the learning status space of complex foraging task and improve the learning speed , a double-deck hierarchical reinforcement learning with share zone is presented . The arithmetic can perform not only the lower hierarchical of state-action learning but also the higher hierachical of station-behavior learning . the higher hierachical of station-behavior learning can avoid the combination explosion of status space . the use of the share zone reinforces the ability of cooperative learning . Simulation results show that the arithmetic can improve the learning speed of robots and satisfy the time need of multi-robot complex foraging task in unknown environment .

Key words: foraging task, multi-robot systems, reinforcement learning, cooperative