基于未知环境状态新定义及知识启发的机器人导航Q学习算.PDF

2014 年 第 23 卷 第 1 期 计 算 机 系 统 应 用 基于未知环境状态新定义及知识启发的机器人导航Q ① 学习算法 童小龙, 姚明海, 张灿淋 (浙江工业大学 信息工程学院, 杭州 310023) 摘 要: 由于强大的自主学习能力, 强化学习方法逐渐成为机器人导航问题的研究热点, 但是复杂的未知环境对 算法的运行效率和收敛速度提出了考验. 提出一种新的机器人导航Q 学习算法, 首先用三个离散的变量来定义环 境状态空间, 然后分别设计了两部分奖赏函数, 结合对导航达到目标有利的知识来启发引导机器人的学习过程. 实验在 Simbad 仿真平台上进行, 结果表明本文提出的算法很好地完成了机器人在未知环境中的导航任务, 收敛 性能也有其优越性. 关键词: 强化学习; 状态定义; 知识启发; Simbad 平台 A Q-Learning Algorithm for Robot Navigation Based on a New Definition of an Unknown Environment States and Knowledge Heuristic TONG Xiao-Long, YAO Ming-Hai, ZHANG Can-Lin (College of Information Engineering, Zhejiang University of Technology, Hangzhou 310023, China) Abstract: Due to powerful self-learning ability, reinforcement learning has become a research hot spot about robot navigation problems, but the operating efficiency and convergence speed of the algorithm are tried by the the complex unknown environment. A new Q-learning algorithm for robot navigation was proposed in this paper. First, three discrete variables were used to define the space states of the environment, and then two parts of the reward functions were designed, combining the beneficial knowledge for reaching the target to inspire and guide the robots learning process. The experiment was executed on Simbad simulation platform. The results show that the proposed algorithm is well done in an unknown environment robot navigation task, and has a better convergence speed. Key words: reinforcement learning; states definition; knowledge heuristic; Simbad platform 未知环境下移动机器人导航主要面临解决两大问 在未知环境中探索的重要技术, 很多传统的人工智能 题, 即机器人定位和路径规划[1,2]. 定位是指机器人根 算法在这方面做了大量的工作. 随着机器人应用领域 据所处的环境判断自己的位置和方向, 它需要识别出