Synchronized Multi-Drone Delivery via Public Transport Mobile Depots: A Dual-Phase Hybrid MILP and RL Framework

Abstract

The integration of Unmanned Aerial Vehicles (UAVs) with Bus Rapid Transit (BRT) networks offers a highly efficient paradigm for urban last-mile delivery. Operationalizing this system requires reconciling rigid timetables, stochastic demand, and non-linear energy consumption. This paper introduces a novel dual-phase optimization framework to address the Stochastic Public Transport-Based Drone Delivery Problem. Phase 1 employs a Mixed-Integer Linear Programming (MILP) model to establish a deterministic baseline schedule for planned parcels, strictly enforcing spatial-temporal synchronization with the mobile depot. Phase 2 introduces a Q-Learning dynamic controller to manage the stochastic arrival of high-yield express parcels. The Markov Decision Process (MDP) uses a custom reward function that balances financial yield against an aerodynamic power model and synchronization penalties. Computational experiments, simulating operations under variable meteorological conditions, demonstrate that the proposed hybrid architecture significantly increases total shift revenue while maintaining a zero-failure rate for drone-bus rendezvous.