基于改进DQN算法的机器人路径规划  

Robot Path Planning Based on Improved DQN Algorithm

作　　者：马浩东 陈玲玲[1] 金小杭 MA Hao-dong;CHEN Ling-ling;JIN Xiao-hang(Jilin Institute of Chemical Technology,Jilin 132000,Jilin)

机构地区：[1]吉林化工学院,吉林吉林132000

出　　处：《电脑与电信》2023年第11期37-41,共5页Computer & Telecommunication

基　　金：吉林省自然科学基金,项目编号:YDZJ202201ZYTS653。

摘　　要：为了解决机器人路径规划到达目标点困难,以及容易碰撞的问题,研究提出一种改进的DQN算法并命名为MDQN,通过改变状态空间的设置,能更准确地区分不同的状态,得到一个更符合实际的观测空间;通过改变奖励函数,能够获得更有价值的奖励值,从而提高学习效率。实验在栅格地图中进行,在Pycharm平台上仿真,结果表明,改进的算法与传统DQN算法相比,具有更高的奖励值、能够顺利到达目标点,且路径更短等优点。To address the issues of slow convergence and low success rate in robot path planning associated with the DQN algorithm,this study introduces an enhanced version of the DQN algorithm named MDQN.This enhancement involves modifying the configuration of the state space to more accurately distinguish different states,resulting in a more realistic observation space.Additionally,by altering the reward function,the algorithm is capable of obtaining more reward values,thereby improving learning efficiency.Experimental evaluations conducted on grid maps demonstrate that the improved algorithm,when compared to the traditional DQN algorithm and other related approaches,offers advantages such as higher success rates,faster convergence,and shorter paths.

关 键 词：路径规划 DQN算法 状态空间 奖励函数 栅格法 

分 类 号：TP242[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]