Collaborative Delivery Optimization With Multiple Drones via Constrained Hybrid Pointer Network
Fanhui Kong, Bin Jiang, Jian Wang, Huihui Wang, Houbing Song
Abstract
Drone participation in truck delivery is a potential booster for the last-mile logistics system, which has been an emerging hot research field. Among that, how to arrange a fleet of drones from the truck and optimize the vehicle routing problem with drones (VRPDs) is a key issue. However, most existing studies fail to derive the feasible solutions due to unordered customer distributions and multivariant drone feature constraints. In this article, we propose a novel self-driven reinforcement learning structure, named constraint-based hybrid pointer network (CH-Ptr-Net) model, which is a hybrid pointer network approach composed of graph neural network (GNN) embedding and attention decoder. We go into developing the simpler embedding version for multiple drones-assisted truck delivery. The CH-Ptr-Net model tends to generate a set of optimal delivery sequence, after constructing the mixed-integer linear program (MILP) formulation. Extensive numerical testing indicates that the proposed method performs better than recent exact and heuristic approaches for collaborative delivery routing optimization with the truck carrying multiple drones.