Resilient Drone-enabled Logistics: Integrating Truck-and-Drone Planning with Disruption Response

Abstract

This dissertation investigates the impact of sudden disruptions such as adverse weather on last mile delivery drone and truck tandems, known as the Flying Sidekick Traveling Salesman Problem (FSTSP). It presents the first extension of the FSTSP that incorporates recourse strategies for disruption management and develops modeling approaches to improve the resilience of truck-and-drone delivery operations. The first approach proposes a two-stage disruption-aware optimization model in which an initial truck-and-drone delivery plan is revised after a disruption occurs. This is a deterministic approach that is fully reactive. Both full region disruption (total) and moving partial region disruption (rolling) are considered. Mixed Integer Linear Programming (MILP) is used to solve small instances, while a heuristic adapted from the literature is developed for larger instances. Compared with post-disruption truck-only operations, the proposed recourse strategy reduces makespan by about ten percent for instances with up to 100 customers while achieving performance close to a perfect-information benchmark. The next approach introduces a dynamic replanning framework for disruptions, where disruption start times and durations are modeled stochastically. A receding horizon algorithm continuously updates disruption information at each decision epoch and re-optimizes the remaining truck-and-drone delivery plan throughout the planning horizon. As with the first approach, this is applied to total and to rolling disruptions. Computational results for the dynamic replanning models demonstrate that incorporating recourse strategies significantly improves delivery performance for both total and rolling disruptions by reducing makespan and increasing the number of customers served by drones iii compared with the corresponding no-recourse strategies. Statistical analyses confirm that disruption start-time and duration distributions have statistically significant effects on delivery performance. For both approaches, rolling disruptions outperform total disruptions since the non-disruption portions of the region can still potentially use drone delivery. Overall, the proposed models provide effective disruption-aware planning strategies and demonstrate the importance of incorporating recourse into truck-and-drone delivery systems for resilient and efficient last-mile logistics. The solution approach is pragmatic both in performance and computational effort even for large problems.