Data Collection with Multiple UAVs via Multi-Agent Reinforcement Learning

Unmanned aerial vehicles (UAVs) have been increasingly incorporated into wireless networking scenarios. In particular, recent studies in the literature have considered the utilization of UAVs for data collection in specific environments via deep reinforcement learning (DRL). However, most studies just address using a single UAV to complete the data collection tasks, potentially leading to high latency or insufficient data collection. In this work, we introduce multiple UAVs to accomplish data collection via multi-agent DRL in the presence of other non-data-collecting UAVs with unknown trajectories. We take into account kinematic and collision avoidance constraints. We also expand our approach to different random environments and evaluate its robustness. We design two different algorithms and show that both can help UAVs establish a stable and efficient policy to accomplish the data-collection task.