A reinforcement learning method to solve dynamic truck-drone routing problem

This paper develops a reinforcement learning method to solve the dynamic routing problem of a truck-and-drone collaboration system with multiple types of tasks. Specifically, the collaboration system considers a post-disaster scenario. On the one hand, trucks and drones can deliver necessary relief logistics (medicine, food, and water) to the affected people. On the other hand, drones, with small cameras onboard, can perform surveillance tasks to assess the network and find new demand after disasters. Trucks and drones can collaborate on serving all demands. As new demands may arrive during the rescue process, the route of trucks and drones should be updated to rescue more people. In this paper, we develop a reinforcement learning method to solve the proposed problem in real time. Since the problem is an NP-hard problem with a large state and action space, the attention mechanism is adopted for the state representation. In numerical studies, we extensively compare the proposed reinforcement learning methods with other methods.