Heterogeneous Attentions for Solving Pickup and Delivery Problem via Deep Reinforcement Learning

Recently, there is an emerging trend to apply deep reinforcement learning solve the vehicle routing problem (VRP), where a learnt policy governs selection of next node for visiting. However, existing methods could not handle well pairing and precedence relationships in pickup delivery (PDP), which representative variant VRP. To address this challenging issue, we leverage novel neural network integrated with heterogeneous attention mechanism empower automatically select nodes. In particular, specifically prescribes attentions each role nodes while taking into account constraint, i.e., must precede node. Further masking scheme, expected find higher-quality solutions solving PDP. Extensive experimental results show that our method outperforms state-of-the-art heuristic model, respectively, generalizes different distributions sizes.