摘要
提出增强学习(RL)解决机器人动态操作任务运动规划的方法。对动态操作任务,分析了如何确定输入输出变量以及强化函数的设计问题;给出用于连续输入输出问题的自适应启发评价(AHC)算法。增强学习解决动态操作任务的运动规划问题,只需要机器人正解进行反复尝试即可学会动作,从而避免了常规运动规划方法中涉及的复杂逆解运算;最后以平面3连杆机器人接取自由飞行的球为例进行仿真研究,结果表明了方法的有效性和可行性。
Reinforcement learning (RL) to motion planning of dynamic manipulation tasks was applied, The input(s), the output(s) and reinforcement function were analyzed, and adaptive heuristic critic (AHC) algorithms were adopted for continuous problem. The advantage of applying RL to dynamic manipulation is to avoid the complex inverse kinemics and to learn the motion by trial. Simulation of planar 3 links manipulator to catch free flying ball is to validate the method.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2006年第9期2537-2540,共4页
Journal of System Simulation
关键词
增强学习
运动规划
动态操作任务
reinforcement learning
motion planning
dynamic manipulation task