MADRL-Based Resource Allocation for UAV-Assisted Mobile Edge Computing
This paper proposes a joint optimization algorithm for trajectory control and task offloading ratios based on multi-agent deep reinforcement learning. By jointly optimizing the flight trajectories of unmanned aerial vehicles (UAVs), user scheduling strategies, and task offloading ratios, the decoupled coordination of resource allocation and trajectory planning is achieved, thereby minimizing system delay and weighted energy consumption. An enhanced multi-agent proximal policy optimization algorithm, named FMAHPPO, is designed. Compared with existing benchmark algorithms, the FMAHPPO algorithm significantly reduces the total system overhead and effectively improves the energy efficiency and task processing success rate of multi-UAV swarms. This research provides a valuable theoretical foundation and algorithmic support for the collaborative management of edge resources in future space-air-ground integrated networks (SAGIN).