A Multi-Objective Drone Routing Problem for On-Demand Delivery Considering Hybrid Delivery Modes

This paper investigates the multi-objective drone routing problem for on-demand delivery considering hybrid delivery modes. Unlike previous studies that assume a single delivery strategy or statically bind modes to heterogeneous drone types, real-world operations require a hybrid framework where homogeneous drones dynamically switch between exclusive and sharing modes to accommodate orders with distinct logics. We formulate a multi-objective mixed-integer programming model that minimizes operational costs while maximizing order revenue, explicitly accounting for dynamic order arrivals, drone battery swapping, and resource conflicts at shared lockers. To solve this problem under dynamic conditions, we propose an online optimization framework, the reinforcement learning knowledge-driven multi-objective evolutionary algorithm based on decomposition (RL-KMD-MOEA/D), which integrates Q-learning for adaptive operator selection within a rolling-horizon scheme. Comprehensive experiments demonstrate that RL-KMD-MOEA/D achieves competitive performance on small-scale instances and exhibits superior scalability and robustness on large-scale, highly constrained dynamic scenarios, outperforming other compared algorithms.