基于DQN算法的农用无人车作业路径规划
Route Planning of Agricultural Unmanned Vehicle Based on DQN Algorithm
庄金炜 1张晓菲 2尹琪东 1陈克1
作者信息
- 1. 沈阳理工大学汽车与交通学院,沈阳 110159
- 2. 中国人民解放军第六四零九工厂研究院,辽宁抚顺 113000
- 折叠
摘要
传统农用无人车作业时常依据人工经验确定作业路线,面对复杂的作业环境时无法保证路径规划的高效性,且传统覆盖路径规划方法聚焦于覆盖率而忽略了车辆作业路线上的损耗.为此,提出一种以减少车辆在路线上的损耗为目标的最优全局覆盖路径规划方法.以深度Q网络(DQN)算法为基础,根据作业时车辆的真实轨迹创建奖励策略(RLP),对车辆在路线上的损耗进行优化,减少车辆的转弯数、掉头数及重复作业面积,设计了 RLP-DQN算法.仿真实验结果表明,对比遗传算法、A*算法等传统路径规划方法,本文RLP-DQN算法综合性能较好,可在实现全覆盖路径规划的同时有效减少路线损耗.
Abstract
In the traditional agricultural unmanned vehicle operation,the route is often determined according to manual experience,which cannot guarantee the efficiency of route planning in the face of complex operating environment.Moreover,the traditional coverage path planning method focuses on the coverage rate,but ignores the loss on the vehicle operation route.Therefore,an optimal glob-al coverage route planning method aiming at reducing the loss of vehicles on the route is proposed.Based on the deep Q network(DQN)algorithm,the reward strategy(RLP)is created according to the real track of the vehicle during operation,and the loss of the vehicle on the route is optimized to reduce the number of turns,U-turns and repeated operation area of the vehicle.Simulation results show that compared with traditional path planning methods such as genetic algorithm and A*algo-rithm,the RLP-DQN algorithm in this paper has better comprehensive performance and can effec-tively reduce route loss while realizing full coverage path planning.
关键词
农用无人车/路径规划/深度强化学习/DQN算法Key words
unmanned agricultural vehicles/path planning/deep reinforcement learning/DQN algorithm引用本文复制引用
基金项目
辽宁省教育厅高等学校基本科研项目(JYTQN2023060)
出版年
2024
NETL
NSTL
万方数据