Deep RL model optimizes truck-drone delivery with lockers and no-fly zones

A team of researchers has tackled a major challenge in last-mile logistics: coordinating trucks and drones while navigating real-world constraints like no-fly zones and battery limits. Their paper, "Locker-based Truck-Drone Routing with Integrated Considerations of Pickups, Deliveries, and No-Fly Zones," presents a new framework called LTDRP-PDNF. Smart lockers serve dual roles – as parcel storage and automated drone docking stations for takeoff, landing, handover, and battery swaps. The system must simultaneously plan deliveries, return pickups, respect battery and load-dependent drone flight ranges, and reroute around restricted airspace.

The proposed solution uses a two-stage deep reinforcement learning neural heuristic. The first stage solves the truck-only routing problem (a capacitated vehicle routing problem) using an attention-based encoder and a Bidirectional Gated Recurrent Unit decoder. The second stage combines a policy-transfer strategy with a hybrid dispatch assignment heuristic to build fully coordinated truck-drone routes. Experiments across various instance scales show that this approach outperforms both traditional metaheuristics and other neural heuristics in most cases while maintaining exceptionally short computation times, making it a scalable and practical solution for real-world deployment.