基于Q-Learning的无人驾驶船舶路径规划  

On Path Planning of Unmanned Ship Based on Q-Learning

作　　者：王程博 张新宇[1] 邹志强[2] 王少博[2] WANG Cheng-bo;ZHANG Xin-yu;ZOU Zhi-qiang;WANG Shao-bo(Key Laboratory of Marine Simulation and Control for Ministry of Communications Dalian Maritime University,Dalian Liaoning 116026,China;Traffic Information Engineering Institute,Dalian Maritime University,Dalian Liaoning 116026,China)

机构地区：[1]大连海事大学航海动态仿真与控制交通行业重点实验室,116026 [2]大连海事大学交通信息工程实验室,116026

出　　处：《船海工程》2018年第5期168-171,共4页Ship & Ocean Engineering

基　　金：国家自然科学基金(51309043);中央高校基本科研业务费专项(3132016321;3132016315);辽宁省高校杰出青年学者成长计划(LJQ2014052);辽宁省教育厅重点实验室项目(LZ2015009)

摘　　要：为实现无人驾驶船舶在未知环境中自适应航行,建立一种基于Q-Learning的无人驾驶船舶路径规划模型。应用基于马尔科夫过程的Q学习算法,分别就环境模型、动作空间、激励函数及动作选择策略4大要素建立模型,设计激励函数,规划最优策略,使得无人驾驶船舶路径规划过程中所获奖赏最大;利用python和pygame平台建立仿真环境,仿真结果表明,该方法可有效地在未知环境中规划出较优路径及成功避让多个障碍物。To achieve unmanned ship safety of navigation in unknown environments,a path planning model of unmanned ships based on Q-Learning was established.The Q-learning algorithm based on Markov process was applied to setup the simulation model according to the environmental model,action space,excitation function and action selection strategy.A reward function was specifically designed for target approaching and safety.The optimal strategy of path planning was determined by making unmanned ships awarded the most.A complex simulation environment is built by Python and Pygame,to verify the effectiveness of path planning for unmanned ships based on Q-Learning algorithm.The simulation results showed that this method can effectively plan the optimal path in an unknown environment,and multiple obstacles can be avoided successfully.

关 键 词：Q-LEARNING 路径规划 避障 无人驾驶船舶 

分 类 号：U664.7[交通运输工程—船舶及航道工程]