Truck-Drone Delivery Optimization Based on Multi-Agent Reinforcement Learning

Abstract

In recent years, the adoption of truck–drone collaborative delivery has emerged as an innovative approach to enhance transportation efficiency and minimize the depletion of human resources. Such a model simultaneously addresses the endurance limitations of drones and the time wastage incurred during the “last-mile” deliveries by trucks. Trucks serve not only as a carrier platform for drones but also as storage hubs and energy sources for these unmanned aerial vehicles. Drawing from the distinctive attributes of truck–drone collaborative delivery, this research has created a multi-drone delivery environment utilizing the MPE library. Furthermore, a spectrum of optimization techniques has been employed to enhance the algorithm’s efficacy within the truck–drone distribution system. Finally, a comparative analysis is conducted with other multi-agent reinforcement learning algorithms within the same environment, thus affirming the rationality of the problem formulation and highlighting the algorithm’s superior performance.

Keywords

• reinforcement learning
• drone
• multi-agent problem
• path planning