Multi-Agent Reinforcement Learning for Adaptive Logistics Scheduling in Global Supply Chains

Abstract

Modern supply chains involve complex coordination among transportation networks, distribution centers, and inventory management systems. Rapid changes in demand, transportation availability, and delivery conditions frequently reduce the effectiveness of traditional logistics scheduling methods. This study introduces a multi-agent reinforcement learning framework for adaptive transportation scheduling and resource allocation in large-scale supply chain networks. The proposed approach models distribution centers and transportation vehicles as cooperative decision-making agents that optimize delivery efficiency under changing operational constraints. The illustrative evaluation incorporates 5.6 million shipment records from 82 distribution centers and 1,450 transportation routes across multiple geographical regions. Simulation-based experiments indicate that the proposed framework reduces average delivery delays by 18.7% and transportation costs by 12.4% compared with a conventional rule-based scheduling system. On-time delivery performance improves from 86.3% to 93.1%, while average vehicle utilization increases by 9.6 percentage points. Additional experiments demonstrate improved scheduling stability under unexpected demand fluctuations and temporary route disruptions. These findings suggest that cooperative reinforcement learning can provide more adaptive and efficient decision support for complex logistics networks.