We study a city-wide two-echelon last-mile delivery system combining conventional trucks and autonomous vehicles to handle more than 10 000 daily demands. The resulting large-scale multi-trip 2E-VRP is decomposed into a multi-trip routing problem at the second echelon and an inventory routing problem at the first echelon.
Our approach relies on a best-fit insertion heuristic enhanced with a labeling strategy and a multi-level filtering system to ensure tractability, followed by an MILP formulation for the replenishment decisions and a tailored Large Neighborhood Search to improve solution quality.