机电工程技术2024，Vol.53Issue(8)：23-27,88,6.DOI:10.3969/j.issn.1009-9492.2024.00089
基于冲突搜索增强深度强化学习的多AGV路径规划方法
Conflict-based Search Enhanced Deep Reinforcement Learning Approach for Multi-AGV Path Planning
摘要
Abstract
To address the path planning problem for automated guided vehicles(AGVs)with the objective of minimizing total travel time,an approach that enhances distributed independent Q-learning(IQL)with the conflict-based search(CBS)algorithm is proposed.Initially,a grid map method is employed to construct the environmental map,and the multi-AGV path planning problem is mathematically described,including assumptions about types of collisions between AGVs.Subsequently,the problem is transformed into a partially observable Markov decision process(POMDP),with detailed definitions of the observation space,action space,and reward function.Further,by utilizing an asynchronous priority experience replay architecture,the IQL method is extended to a distributed environment,and guided by the CBS algorithm to refine the Q-network,enhancing decision-making processes for AGVs in congested environments.Finally,comparative experiments with other deep reinforcement learning algorithms are designed based on varying numbers of AGVs.The results demonstrate that the proposed method outperforms the control algorithms in key performance indicators such as success rate and average step length,thereby validating the effectiveness and feasibility of the proposed approach.关键词
基于冲突搜索/深度强化学习/多AGV/路径规划Key words
conflict-based search/deep reinforcement learning/multi-AGV/path planning分类
信息技术与安全科学引用本文复制引用
王亦晨,刘雪梅..基于冲突搜索增强深度强化学习的多AGV路径规划方法[J].机电工程技术,2024,53(8):23-27,88,6.基金项目
西门子(中国)有限公司合作项目(kz0100020210591) （中国）
