未知环境中基于强化学习的移动机器人路径规划.pdfVIP

第29 卷第4 期 机 电 工 程 Vol. 29 No.4 A-PDF Split DEMO : Purchase from www.A-PDF.com to remove the watermark 2012 年4 月 Journal of Mechanical Electrical Engineering Apr. 2012 未知环境中基于强化学习的 移动机器人路径规划* 梁 泉 （南京农业大学 工学院，江苏 南京 210031） 摘要：为解决未知环境中移动机器人的自适应路径规划问题，提出了一种基于Q 学习算法的自主学习方法。首先设计了未知环境中基 于传感器信息的移动机器人自主路径规划的学习框架，并建立了学习算法中各要素的数学模型；然后利用模糊逻辑方法解决了连续状 态空间的泛化问题，有效地降低了Q值表的维数，加快了算法的学习速度；最后在不同障碍环境中对基于Q 学习算法的自主学习方法进 行了仿真实验，仿真实验中移动机器人通过自主学习较好地完成了自适应路径规划。研究结果证明了该自主学习方法的有效性。 关键词：未知环境；Q 学习算法；移动机器人；路径规划 中图分类号：TP242 文献标志码：A 文章编号：1001-4551 （2012 ）04-0477-05 Reinforcementlearningbasedmobilerobotpathplanning inunknownenvironment LIANG Quan （College of Engineering ，Nanjing Agricultural University ，Nanjing 210031 ，China） Abstract：In order to solve problem of the adaptive path planning on mobile robot in unknown environments ，a self-learning method based on Q-learning algorithm was proposed. Firstly ，the learning framework was designed for the adaptive path planning of mobile robot in unknown environments based on sensor information ，and the mathematics model for each element of learning algorithm was proposed. Then the generalization problem of continuous state space of reinforcement learning system was solved by fuzzy logic ，the size of the Q-table was reduced and the speed of the learning algorithm was increased. Finally ，the simulation of self-learning method based on Q-learnning algorithm was carried in the environment with different obstacles ，the adaptive path planning was achieved by the mobile robot thr