Reinforcement Learning for Practical Express Systems with Mixed Deliveries and Pickups

In real-world express systems, couriers need to satisfy not only the delivery demands but also pick-up of customers. Delivery and pickup tasks are usually mixed together within integrated routing plans. Such a problem can be abstracted formulated as Vehicle Routing Problem with Mixed Pickup (VRPMDP), which is an NP-hard combinatorial optimization problem. To solve VRPMDP, there three major challenges below. (a) Even though successive independent accomplish, inter-influence between choosing task or deal still exists. (b) Due two-way flow goods depot customers, loading rate vehicles leaving affects decisions. (c) The proportion deliveries pickups will change due complex demand situation in scenarios, requires robustness algorithm. above, we design encoder-decoder based framework generate high-quality robust VRPMDP solutions. First, consider instance graph utilize GNN encoder extract feature effectively. detailed solutions further decoded sequence by decoder attention mechanism. Second, propose Coordinated Decision Loading (CDLR) mechanism determine dynamically after vehicle returns depot, thus avoiding influence improper settings. Finally, model equipped CDLR simultaneously adapt changes pickups. We conduct experiments demonstrate effectiveness our model. show that method achieves desirable results generalization ability.