Multi-UAV Reinforcement Learning for Data Collection in Cellular MIMO Networks

Uncrewed Aerial Vehicles (UAVs) provide a compelling solution for data collection in Internet of Things (IoT) networks due to their mobility and adaptability. However, the line-of-sight dominance in their channels may result in severe interference to ground users during UAV operations. To address this, we present an optimization framework that concurrently optimizes UAV trajectories and transmit powers. Our approach efficiently results in the collection of data from a variety of IoT sensors while (a) minimizing the UAVs flying time and (b) mitigating interference with terrestrial networks. Given the complex nature of such an optimization problem, this paper leverages reinforcement learning, specifically the twin delayed deep deterministic policy gradient algorithm, where a distributed learning algorithm is presented. Experimental results validate the efficacy of our proposed approach, demonstrating its capability to significantly enhance data collection in IoT networks while minimizing UAV flight time and interference with ground user links.