Dynamic Route Planning for Last-Mile Delivery

Abstract

There has never been a time with more demand than now for e-retailing and as a consequence last-mile services. The growth in demand is also bringing significant challenges. With the abundance of options, customers are ever more demanding and expecting more control. With the existing strategies, matching customers' foregoing expectations causes significant economic burdens and ecological disturbances. As a result, e-retailers need to define efficient routing strategies for their last-mile services. This thesis is motivated by identifying efficient routing strategies, in terms of environmental impacts, service time and cost, for last-mile delivery services. We investigate different routing strategies for the last-mile delivery problems, with a focus on same-day services. The corresponding problem is known as the last-mile same-day delivery problem and is dynamic due to the nature of service requests.

In the first part, we investigate vehicle and drone integrated delivery systems. We consider an alternative way to integrate drones into conventional vehicle delivery systems, such that drones resupply vehicles with the future orders of customers while vehicles deliver available orders to customers. We evaluate the impact of the drone resupply system based on a case of the problem in which a single vehicle and a single drone are dedicated to the service area. We introduce a mixed-integer programming model for the delivery problem with known requests. For the dynamic problem in which the requests reveal dynamically throughout the horizon, we propose a periodic reoptimization algorithm as a solution approach. We compare the performance of the drone resupply system to the conventional vehicle only delivery systems over several practical instances that differ in terms of customer preferences and system settings. Through computational experiments, we showed that the drone resupply system outperforms the conventional system with respect to operational time, cost and carbon emissions levels.

In the second part of the thesis, we evaluate the impact of outsourcing strategy in a multi-period delivery problem. Given the relevance of the problem in practice, we suggest that exploitable stochastic information might be gathered for the dynamically revealed information. To the best of our knowledge, we are the first to introduce outsourcing in the literature of dynamic multi-period vehicle routing problems with probabilistic information. We model the corresponding problem as a Markov decision process. We propose a multi-stage programming model and a progressive hedging algorithm to solve the decision problems. We evaluate several planning strategies to evaluate the impact of postponement and outsourcing decisions. Based on the computational experiments, we determined the best delivery strategy in terms of cost over different practical settings.