基于多智能体深度强化学习的机器人协作搬运方法

The robot cooperative handling method based on multi-agent deep reinforcement learning

近年来,机器人协作搬运任务在生产线和无人仓库场景中得到广泛应用。针对在传统路径规划方法上机器人无法达到最高的搬运效率的问题,提出一种通过基于多智能体深度强化学习(Multi-Agent Deep Reinforcement Learning,MADRL)的QTRAN Plus算法参与协作搬运的机器人规划路径。QTRAN Plus算法采用混合网络代替QTRAN算法中对每个智能体的Q值网络进行加和的操作,以提高优化能力,并增加了一个新的损失函数,以提高收敛速度。通过协作搬运仿真实验可知,QTRAN Plus能够更快更稳定地学习到机器人的最优路径,其整体表现优于其他对比算法。

In recent years,the cooperative robot handling tasks have been widely used in production lines and unmanned warehouse scenarios.The robots may not achieve the highest efficiency with the traditional path planning method.To this end,the QTRAN Plus algorithm based on Multi-Agent Deep Reinforcement Learning is proposed in this paper to plan the path of the robot involved in cooperative handling.The QTRAN Plus algorithm uses hybrid network instead of QTRAN algorithm to add the Q-value network of each agent to improve the optimization ability,and adds a new loss function to improve the convergence speed.The simulation results show that QTRAN Plus can learn the optimal path of the robot faster and stably,and its overall performance is better than other comparison algorithms.