多障碍环境中基于增强式学习的势场优化和机器人路径规划  

Potential Field Optimization and Robot Path Planning in Multi-Obstacle Environment Based on Reinforcement Learning

作　　者：庄晓东[1] 孟庆春[1] 王汉萍[1] 殷波[1] 

机构地区：[1]青岛海洋大学电子工程系,青岛266003

出　　处：《青岛海洋大学学报（自然科学版）》2001年第6期937-942,共6页Journal of Ocean University of Qingdao

基　　金：高等学校重点实验室访问学者基金;青岛市科委课题资助

摘　　要：该文把增强式学习方法应用于多障碍环境中机器人路径规划 ,并将增强式学习和路径规划相结合 ,通过工作空间势场的自适应优化学习 ,实现机器人的全局路径规划 ,即得到从任何初始位置开始的最优路径。与传统的人工势场方法相比 ,该方法避免了势场中局部极小点所引起的陷阱区域 ,并且所得到的路径具有最优特性。计算机仿真实验结果表明 。In this paper the reinforcement learning to robot path planning in complex environment of multiple obstacles is applied. An adaptive control strategy learning method is proposed. Reinforcement learning is an unsupervised learning method based on the reactive and feedback mechanism. In this paper the reinforcement learning and path planning are combined together. The optimal path from any initial position is obtained by optimizing the global potential field and control rules. Compared with traditional artificial potential field method, this method avoids irrelevant local minimal points, which can make the robot vibrate in a small local area. Furthermore, the path found is optimal. The computer simulation experiment result shows that this learning method can efficiently solve the robot path planning problem in multi obstacle environment.

关 键 词：增强式学习 移动机器人 多障碍环境 人工势场 路径规划 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置] TP18[自动化与计算机技术—控制科学与工程]