Joint Trajectory and Beamforming Optimization for UAV-Enabled SAGIN via MADRL

In this paper, we construct a space-air-ground integrated network (SAGIN) composed of satellite, unmanned aerial vehicles (UAVs), and internet of things (IoT) devices for information collection. The IoT devices collect environmental information, while UAVs equipped with an integrated sensing and communication (ISAC) system, perform location sensing of the IoT devices. Based on the acquired location, the UAVs dynamically plan their data collection trajectories. To achieve a balance between communication and sensing performance, a multi-agent deep reinforcement learning (MADRL) framework is adopted to jointly optimize the UAV trajectories, matching, and beamforming strategies. The satellite, acting as the central node, conducts offline learning based on the experiences of all UAVs to obtain the optimal policy, and then distributes the learned policy and state value functions to each UAV. Simulation results demonstrate that the proposed method offers significant advantages over existing solutions in improving system performance and reducing complexity.