Rolling-horizon replanning for two-echelon truck–drone routing with dynamically released pickups

In urban last-mile delivery and pickup operations, trucks transport scheduled delivery parcels to staffed satellites, where drones handle customer deliveries and pickups. Payload-dependent energy use, customer time windows, satellite loading, and dynamically released pickups jointly determine route feasibility. We formulate an online rolling-horizon model that preserves completed or operationally committed actions and inherits the current vehicle states at each decision epoch. We further develop a rolling-horizon route-pool optimization framework (RH-RPO) that combines feasible route-pool construction, route-based mixed-integer optimization, and execution-aware online updates of the still-adjustable drone operations. Any stage-level certificate is explicitly limited to the corresponding route-master model and enumerated customer-order space and does not imply full-horizon optimality over unrevealed future requests. Computational experiments on 45 dynamic instances with 15–50 customers show that RH-RPO returns executable plans in all cases. The Stage-0 route master is solved to certified optimality under its canonical-departure definition in 41 instances. Separately, RH-RPO obtains a lower evaluated Stage-0 operating cost than each benchmark in all 45 paired instances. At the same time, its later-stage replanning time remains below 0.004 s on average. Most computation occurs at Stage-0, and later updates are lighter because fewer operations remain adjustable based on the previously frozen execution boundary.