基于强化学习的多机器人避碰算法研究

Algorithm of multi-robot collision avoidance based on reinforcement learning

采用强化学习解决多机器人避碰问题。然后针对表格式Q学习算法只能用于离散的状态并且学习时间过长,难以收敛的不足,提出了神经网络和Q学习相结合的算法。最后将该算法应用到多机器人避碰问题中,仿真实验表明该算法有效,能较好地解决多机器人避碰问题。

This paper adopts reinforcement learning to solve multi-robot collision avoidance problems. Then in allusion to the insufficiency that tabular Q-learning algorithm can only be used for discrete states and learning time is too long, difficult to convergence, it puts forward combination of neural networks and Q-learning algorithms. Finally the algorithm is applied to multi-robot collision avoidance problems. The simulation experiments show that the algorithm is effective and well solve the multi-robot collision avoidance problems.