Joint Optimization of 3-D UAV Trajectory and Task Scheduling With Multi-Agent Deep Reinforcement Learning

Unmanned Aerial Vehicle (UAV)-assisted Mobile Edge Computing (MEC) has emerged as a promising paradigm for supporting computation-intensive applications of ground User Equipments (UEs). However, with the growing computational demands for UEs, it becomes crucial and extremely challenging to jointly optimize the Three-Dimensional (3D) UAV trajectory and task scheduling decision, given UAVs’ limited coverage and scarce resources. To address these challenges, we develop a Multi-UAV Trajectory Optimization and Task Scheduling (MUTOTS) algorithm based on Multi-Agent Deep Reinforcement Learning (MADRL). The proposed MUTOTS framework jointly optimizes multi-UAV 3D trajectories and task scheduling decisions, with the objective of maximizing total system throughput under strict flight-safety constraints. To further improve training stability and convergence, MUTOTS integrates prioritized experience replay and L2 regularization technologies. Moreover, MUTOTS can adapt to different take-off locations, and effectively guide UAVs approach near-optimal deployment locations during the execution stage. Extensive simulation results demonstrate that the effectiveness, convergence and stability of the MUTOTS algorithm, and it significantly improves the system throughput compared with the other existing benchmark schemes.