Coordinated Multi-regional Logistics Path Planning: A Broad Reinforcement Learning Framework

Shengwei Li, Congcong Zhu, Zeping Tong

Abstract: In the context of coordinated multi-regional logistics, a sophisticated path planning approach is essential for optimizing operational efficiency. The dynamic path planning problem, characterized by its complexity and the need for real-time decision-making, presents a significant challenge. Traditional methods, often relying on heuristics, can fall short in providing the most effective solutions, particularly when faced with large-scale data sets. To address these challenges, we introduce a novel framework that employs broad reinforcement learning for coordinated multi-regional logistics path planning. Our algorithm is fortified by a pre-training phase, which enhances its adaptability across diverse scenarios and enables it to swiftly identify near-optimal solutions even with modest data sets. Through extensive experimentation with enterprise and Solomon data sets, our framework has demonstrated superior performance over established algorithms such as MAVFA, MADQN, and NSGA. While computationally intensive compared to heuristic methods, especially with an increase in order volume, our approach consistently outperforms in scenarios prioritizing optimal solution discovery. The implications of this research extend to a wide array of multi-agent sequential decision-making problems, suggesting potential for future exploration and application of our framework in various coordinated logistics endeavors.

External IDs:dblp:conf/ica3pp/LiZT24