MAPDP: Cooperative Multi-Agent Reinforcement Learning to Solve Pickup and Delivery Problems

Cooperative Pickup and Delivery Problem (PDP), as a variant of the typical Vehicle Routing Problems (VRP), is an important formulation in many real-world applications, such on-demand delivery, industrial warehousing, etc. It great importance to efficiently provide high-quality solutions cooperative PDP. However, it not trivial effective directly due two major challenges: 1) structural dependency between pickup delivery pairs require explicit modeling representation. 2) cooperation different vehicles highly related solution exploration difficult model. In this paper, we propose novel multi-agent reinforcement learning based framework solve PDP (MAPDP). First, design paired context embedding well measure nodes considering their limits. Second, utilize decoders leverage decision dependence among vehicle agents on special communication embedding. Third, A2C algorithm train integrated We conduct extensive experiments randomly generated dataset dataset. Experiments result shown that proposed MAPDP outperform all other baselines by at least 1.64\% settings, shows significant computation speed during inference.