基于强化学习的多机器人协作

Multi-Robot Cooperation Based on Reinforcement Learning

提出了一种动态环境下多个机器人获取合作行为的强化学习方法,该方法采用基于瞬时奖励的Q-学习完成单个机器人的学习,并利用人工势场法的思想确定不同机器人的学习顺序,在此基础上采用交替学习来完成多机器人的学习过程。试验结果表明所提方法的可行性和有效性。

In this paper,a kind of reinforcement learning method is addressed to obtain cooperative behaviors for multiple robots under dynamic environment.The proposed learning method adopts Q-learning with immediate reward,and determines learning order for different robots by virtue of the idea of artificial potential field.Based on these modifications,an alternative learning process is introduced to finish the learning process among multiple robots.Experimental results have shown the feasibility and validity of the given approach.