改进的Q学习算法在轨迹规划中的应用  

Improved Algorithm of Q-Learning for Trajectory Planning

作　　者：赵辉[1] 刘雅喆[2] 

机构地区：[1]渤海大学工学院,辽宁锦州121013 [2]大庆师范学院计算机科学与信息技术学院,黑龙江大庆163318

出　　处：《吉林大学学报（信息科学版）》2016年第5期697-702,共6页Journal of Jilin University（Information Science Edition）

基　　金：国家青年基金资助项目(61304053)

摘　　要：为解决Q学习算法易陷入局部最优解问题,改进了传统贪婪策略,提出了一种分段渐近搜索策略。该策略通过动态调整策略参数,使Q学习算法在学习过程中实现探索-学习-利用3个阶段的渐近跳转。同时将该搜索策略应用于Q学习算法中,使改进的Q学习算法能更快速地逼近全局最优解。将改进算法应用于机械臂轨迹规划中,其仿真结果表明,该算法能稳定地引导机械臂沿最优轨迹快速到达目标位置。Aiming at the local optimal was proposed base on greedy strategy. solution for Q learning algorithm, a segment incremental search strategy The improved Q learning jump gradually between three situations such as explore, learn and utilize by adjusting parameters of segment incremental search strategy, and it could approach the global optimal rapidly than the traditional one when the new search strategy is applied to the Q learning. The simulation results show that the manipulator reaches the target position accurately and quickly guided by the improve Q learning algorithm.

关 键 词：在线学习 轨迹规划 机械臂 数学模型 搜索策略 

分 类 号：TP242.2[自动化与计算机技术—检测技术与自动化装置]