MARLDRP: Benchmarking Cooperative Multi-agent Reinforcement Learning Algorithms for Drone Routing Problems

Shiyao Ding, Hideki Aoyama, Donghui Lin

Abstract: The use of drones as an efficient delivery solution is a promising technology, addressing the growing demand for deliveries. Unlike the traditional vehicle routing problem (VRP), we introduce a new drone routing problem (DRP) that considers distinct drone delivery attributes, especially the need for dynamic, collision-free routes in non-grid settings. To optimize team rewards in DRP, cooperative efforts of all drones are essential. Thus, we employ cooperative multi-agent reinforcement learning (MARL). We present MARL\(_{4}DRP\), a comprehensive benchmark tailored for applying cooperative MARL to DRP. Our contributes to the optimization of drone delivery using MARL, offering a solid foundation for future research in this domain. All code is available at the repository: https://github.com/DING-1994/MARL4DRP