Autonomous Vehicles Delivery Systems: Analyzing Vehicle Routing Problems with a Moving Depot

Abstract

The vast growth in the e-commerce market has increased the attention to resolving the problem of Last Mile Delivery that has significant challenges such as reducing operational cost or ecological impact and increasing supply chain performance. The inclusion of new technologies such as drones and robots help tackle these challenges by developing new distribution systems to improve from traditional deliveries methods. However, the use of these technologies brings new operational challenges. This research deals with the impact of using autonomous vehicles in logistics. We first present a technological review of the use autonomous vehicles in logistics and use it to introduce a classification of the delivery systems based on the parcel handover at the time of the last handling before delivery to customers. We describe three categories of handovers, namely, machine-to-person, machine-to-machine, and person-to-machine, and characterize for each of them the type of vehicle routing optimization that it implies. Moreover, we study a truck-drone system, where the truck serves as a depot from where we load the product to the drone for final delivery to customers. The depot is now moving unlike in a traditional Vehicle Routing Problem for which we always assume a fixed depot. Therefore, we present a new class of Vehicle Routing Problems with a moving depot for a truck-drone system and formulate six Integer Linear Programming formulations to minimize the total operational cost through sequencing the deliveries to different customers and optimizing the locations for the truck to release and collect the drones. The problem is NP-hard, thus developing heuristic solutions is more appropriate for large size instances. The proposed models are first solved using the General Algebraic Modeling System software to find the optimal solutions and study their characteristics. Furthermore, a Clarke and Wright Savings heuristic is developed using C++ language to solve large-size problems. The algorithm returned solutions that are within the known quality, 20%. The solutions provided 8% to 20% deviation from the optimal solutions. The algorithm returned solutions for 80 nodes within 1200 seconds. Different real-life applications can adopt the proposed models such as the UPS truck-drone and Amazon airborne fulfilment centre.