”In the Air” ML-driven Optimization of UAV Coverage and Resource Utilization

Unmanned Aerial Vehicles (UAVs) are being deeply investigated because of the several services and improvements they can provide in different application fields. They have proved to be effective especially when the ground users require them to complete specific tasks, such as target tracking, event identification or areas monitoring. Although several works have addressed numerous challenges in using UAVs, their limited time of flight, their trajectories definition and their constrained onboard computation capabilities still need a deeper investigation. In this work, we consider a scenario where ground users assign tasks to UAVs, each task requiring a certain computational effort. The aim is thus to envision a distributed control solution for the UAVs trajectories, which jointly maximizes the coverage of ground users and the utilization of the computational resource utilization of the individual UAVs, while maintaining adequate connectivity between them. A Distributed Deep Reinforcement Learning (DDRL) approach is used for the drones’ trajectory definition, which is based on the knowledge of the positions of the UAV, the positions of the covered devices and their level of activity. The performance evaluation based on some preliminary simulations presents a remarkable increase in terms of coverage and resource utilization of the UAV network. These results show the potentials of the proposed approach and encourage to look into it for defining further improvements.