摘要
针对现有机械臂动态目标抓取方法轨迹规划困难、实时性不足、难以实现六自由度抓取等问题,提出一种基于深度强化学习(deep reinforcement learning,DRL)的机械臂动态目标抓取方法。进行马尔可夫决策过程(Markov decision process,MDP)建模,设计状态空间、动作空间以及奖励函数,实现机械臂对动态目标的六自由度抓取。基于Pybullet构建机械臂动态目标抓取仿真试验环境,对该方法进行训练,将训练得到的策略在新颖场景进行测试,并与经典规划控制的动态目标抓取方法进行对比。仿真结果表明:该方法能实现机械臂对动态目标的六自由度抓取,在抓取成功率和速度上具有优势。
Aiming at the problems of trajectory planning difficulty,insufficient real-time performance and difficulty in realizing six-degree-of-freedom grasping of existing manipulator dynamic target grasping methods,a manipulator dynamic target grasping method based on deep reinforcement learning(DRL)is proposed.The Markov decision process(MDP)is modeled,and the state space,action space and reward function are designed to realize the six-degree-of-freedom grasping of the dynamic target by the manipulator.Based on Pybullet,the dynamic target grasping simulation test environment of manipulator is constructed,and the method is trained.The trained strategy is tested in a novel scene,and compared with the dynamic target grasping method of classical planning control.The simulation results show that the method can realize the six-degree-of-freedom grasping of the dynamic target by the manipulator,and has advantages in grasping success rate and speed.
作者
张轩
卢惠民
任君凯
莫新民
肖浩然
张伟杰
杨璇
Zhang Xuan;Lu Huimin;Ren Junkai;Mo Xinmin;Xiao Haoran;Zhang Weijie;Yang Xuan(Human Enhancement Technology Innovation Center,Northwest Institute of Mechanical&Electrical Engineering,Xianyang 712099,China;College of Intelligence Science and Technology,National University of Defense Technology,Changsha 410073,China)
出处
《兵工自动化》
北大核心
2024年第6期91-96,共6页
Ordnance Industry Automation
关键词
动态目标抓取
马尔科夫
轨迹规划
深度强化学习
六自由度抓取
dynamic target grasping
Markov
trajectory planning
deep reinforcement learning
six-degree-of-freedom grasping
