Multi-agent SAC approach aided joint trajectory and power optimization for multi-UAV assisted wireless networks with safety constraints^☆

Unmanned Aerial Vehicles (UAVs) have demonstrated significant potential as Aerial Base Stations (A-BSs) for providing data services to Ground Users (GUs), attributed to their flexibility, cost-effectiveness, and high likelihood of establishing line-of-sight links. In this article, we formulate the joint power and trajectory optimization problem for a multi-UAV assisted wireless network with no-fly zones constrained, aiming at maximizing the Accumulated Service Data (ASD) of UAVs and minimizing the Average End Age of Information (AEAoI) of GUs. Specifically, this paper proposes the Multi-Agent worst-case Soft Actor Critic (MA-wcSAC) algorithm with a distributional safety-critic. The simulation results demonstrate that, compared to the Multi-Agent Soft Actor Critic (MA-SAC) algorithm, the proposed algorithm exhibits comparable data service performance while reducing security risks by at least 30% at different risk levels.