仓储中基于多智能体深度强化学习的多AGV路径规划

Multi-AGV Path Planning in Warehousing Based on Multi-Agent Deep Reinforcement Learning

随着工业自动化和物流行业的迅速发展,自动引导车辆(Automated Guided Vehicle, AGV)在物流仓库中的路径规划已成为确保运输效率和准确性的关键环节。尽管近年来已经有很多策略被提出,但多AGV系统在复杂的物流环境中仍然频繁地出现碰撞、路径冲突以及控制迟延等问题。鉴于此,本研究提出了一种基于多智能体深度强化学习(Multi Agent Deep Reinforcement Learning, MADRL)的路径规划方法,以期解决多AGV之间的相互协调问题并提高其路径规划效率。为验证所提方法的有效性,我们采用了与遗传算法(Genetic Algorithm, GA)的比较实验。结果显示,基于MADRL的策略在整体运输效率上实现了28%的提升,并在碰撞事件上有了明显的减少。

With the rapid advancement of industrial automation and the logistics industry, the path planning of Automated Guided Vehicles (AGV) in logistics warehouses has become a critical component to ensure transportation efficiency and accuracy. Although numerous strategies have been proposed in recent years, multi-AGV systems still frequently encounter collisions, path conflicts, and control latencies in complex logistics environments. In light of this, our study introduces a path planning approach based on Multi-Agent Deep Reinforcement Learning (MADRL) aiming to address the coor-dination issues among multiple AGVs and to enhance their path planning efficiency. To validate the effectiveness of the proposed method, we conducted comparative experiments with the Genetic Al-gorithm (GA). Results show that the MADRL-based strategy achieved a 28% improvement in overall transportation efficiency and a significant reduction in collision incidents.