面向空间自主装配验证评估的机械臂避障运动规划  

Obstacle Avoidance Motion Planning of Manipulator for Space Autonomous Assembly Validation and Evaluation

作　　者：谢芳霖 汪凌昕 张亚航[1] 王耀兵[1] 王捷[1] XIE Fanglin;WANG Lingxin;ZHANG Yahang;WANG Yaobing;WANG Jie(Beijing Key Laboratory of Spatial Intelligent Robot System Technology and Application,Beijing Institute of Spacecraft System Engineering,Beijing 100094,China)

机构地区：[1]北京空间飞行器总体设计部空间智能机器人系统技术与应用北京市重点实验室,北京100094

出　　处：《航天器工程》2025年第2期82-89,共8页Spacecraft Engineering

摘　　要：面向空间机械臂自主装配的验证评估问题,针对人机、多机协同作业过程中彼此进入对方作业区域后可能出现的高风险碰撞工况,将机械臂避障抓取的运动规划定义为强化学习问题,并提出一种改进的双经验池组合采样经验回放方法。仿真结果表明:采用该方法进行训练后,机械臂可有效避开动态障碍物,且末端到位精度从0.2717 m提高至0.0413 m。基于物理样机开展了试验验证,进一步证明了避障抓取策略能使机械臂自主避开任意障碍物并准确抓取目标,不需要对障碍物的运动状态进行预测。文章提出的避障运动规划,可用于各类固定基座空间机械臂的地面验证评估。The verification and evaluation associated with autonomous assembly using space manipulators is centered on in this paper.Aiming at the high-risk collision scenarios that may arise when astronauts or robots enter each other’s workspaces during human-robot or multi-robot collaborative operations,the motion planning for obstacle avoidance and grasping of the manipulator is formulated as a reinforcement learning problem.Additionally,a method of improved dual-buffer combined experience replay is proposed.After training with this method,the simulations demonstrate that the manipulator can avoid dynamic obstacles,and the end-effector positioning accuracy is improved from 0.2717m to 0.0413m.Experiments based on physical prototypes further verify that the obstacle avoidance grasping strategy enables the manipulator to autonomously avoid any obstacles and precisely grasp target without the prediction for the motion state of obstacles.The obstacle avoidance motion planning proposed in this paper is applicable to ground-based verification and evaluation of various fixed-base space manipulators.

关 键 词：空间机械臂 避障运动规划 深度强化学习 优先经验回放 迁移学习 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置]