【作者】 涂自然； 王维； 梁以业； 禹建丽；

【Author】 Tu Ziran Wang Wei Liang Yiye YU Jianli Department of Mathematics and Physics, He Nan University of Science & Technology Luoyang, 471003; Institute of Electrification & information Engineering,He Nan University of Science & Technology, Luoyang,471003

【机构】 河南科技大学数理系； 河南科技大学电子信息工程学院；

【摘要】 强化学习一词源于行为科学,它模仿人与动物的自然学习过程,通过对环境的反复试探,从而建立从环境状态到行为动作的映射。本文针对基于神经网络结构的机器人全局路径规划算法,利用强化学习的思想, 引进评价预测学习的自适应变步长算法,实现了步长的自动调节,并且加快了路径规划的计算速度,通过仿真试验,表明了所提算法的有效性。更多还原

【Abstract】 Reinforcement learning is an important class of learning techniques that learns to perform a certain task through trial and error interactions with an knowledge-poor environment. This paper studies the problem of adaptive variable stepsize of robotic path planning. The proposed method here using critic prediction learning penalty allows to perform on-line adaptive variance of stepsizes, and the convergence speed of path planning is at least increased bv 10 times more than former’s.更多还原