Coordinating Aerial Transshipment with Ground Delivery for Low-Carbon Urban Logistics

Growing urban last-mile delivery demand and low-carbon transition pressures motivate the coordinated use of unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs). This study develops a two-echelon collaborative delivery model with intermediate transshipment, jointly considering customer allocation, UAV transshipment missions, and UGV routing. A weighted single-objective formulation integrates operational cost, monetized aggregate delivery time, and carbon-emission cost under capacity, UAV-range, and time-window constraints. A learning-enhanced adaptive large neighborhood search method, termed DQN–ALNS, is proposed. A standard deep Q-network with prioritized experience replay selects destroy–repair operator combinations according to the current search state, while ALNS executes these combinations to improve candidate solutions. Experiments cover nine Solomon-based instances with 25, 50, and 100 customers, with ten independent runs per method and instance. DQN–ALNS achieves lower mean objective values than standalone ALNS on six instances, including five of the six medium- and large-scale instances, indicating competitive but instance-dependent performance. Ablation results suggest potential benefits from prioritized experience replay in solution quality and run-to-run stability. A real-data-based scenario study around Nanjing South Railway Station examines different traffic periods. The off-peak scenario yields the lowest integrated objective value of RMB 3042.543 and the shortest delivery completion time of 1.6 h. The proposed framework supports offline route planning and scenario comparison, with DQN–ALNS positioned as a competitive learning-enhanced variant of ALNS.