Hybrid Simulation-Optimization Model To Design Autonomous Delivery Vehicles Networks

This paper proposes a novel integration of simulation, machine learning and mathematical optimization to design a delivery network of autonomous delivery vehicles (ADVs). To obtain the desired scalability, the model is pre-solved using k-means clustering to batch orders based on proximity, then Capacitated Vehicle Routing Problem (CVRP) and Facility Location (FL) models are used to minimize the ADVs’ total traveled distance. The model considers concepts to handle a semi-connected network and use of modular dispenser. Simulation modeling was used to address stochasticity caused by teleoperators required in the process and help identify the resources required for various volume scenarios. A case study is presented where the objective is to optimize the number of ADVs required and teleoperators’ utilization given a stochastic demand.