A multi-agent deep reinforcement learning approach for solving the multi-depot vehicle routing problem

The multi-depot vehicle routing problem (MDVRP) is one of the most essential and useful variants of the traditional vehicle routing problem (VRP) in supply chain management (SCM) and logistics studies. Many supply chains (SC) choose the joint distribution of multiple depots to cut transportation costs and delivery times. However, the ability to deliver quality and fast solutions for MDVRP remains a challenging task. Traditional optimization approaches in operation research (OR) may not be practical to solve MDVRP in real-time. With the latest developments in artificial intelligence (AI), it becomes feasible to apply deep reinforcement learning (DRL) for solving combinatorial routing problems. This paper proposes a new multi-agent deep reinforcement learning (MADRL) model to solve MDVRP. Extensive experiments are conducted to evaluate the performance of the proposed approach. Results show that the developed MADRL model can rapidly capture relative information embedded in graphs and effectively produce quality solutions in real-time.