基于安全强化学习的机械臂无碰路径规划

Collision-Free Path Planning of Manipulator Based on Safety Reinforcement Learning

摘要: 可以通过为机器人关节添加相关约束的方式来保证规划路径的可靠性。本文研究了带有运动学约束的强化学习(Reinforcement Learning)方法，以保证规划中的安全性。通过与替代动作思想的结合，对强化学习动作空间的进行设计，可以进一步保证动作的可行性。为了评估算法性能，在船舶焊接场景中对工业机械臂进行了路径规划，使得机械臂末端成功运动到位于狭窄空间中的焊接起点。实验结果表明，该方法不仅保证了训练的收敛性，而且保证了任务的安全性和可靠性。

Abstract: The reliability of the planned path can be guaranteed by adding relevant constraints to the robot joints. In this paper, reinforcement learning (RL) with motion constraints is studied to ensure the safety in planning. With the design of the reinforcement of the learning movement space by combining with the idea of alternative movements, the feasibility of action is further guaranteed. In order to evaluate the performance of the algorithm, the path planning of the industrial robot arm was carried out in the Marine welding scene, so that the end of the robot arm successfully moved to the welding starting point located in the narrow space. Experimental results show that the proposed method not only guarantees the convergence of the training, but also ensures the security and reliability of the task.