Distributed Connectivity-maintenance Control of a Team of Unmanned Aerial Vehicles using Supervised Deep Learning

This paper presents the design of a decentralized connectivity-maintenance technique for the teleoperation of a team of multiple Unmanned Aerial Vehicles (UAV). Different from current, typical connectivity-maintenance approaches, the proposed technique uses machine learning to attain significantly more scalability in terms of number of UAVs that can be part of the robotic team. It uses Supervised Deep Learning (SDL) with Artificial Neural Networks (ANN), so that each robot can extrapolate the necessary actions for keeping the team connected in one computation step, regardless the size of the team. We compared the performance of our proposed approach vs. a state-of-the-art model-based connectivity-maintenance algorithm when managing a team composed of two, four, six, and ten aerial mobile robots. Results show that our approach can keep the computational cost almost constant as the number of drones increases, reducing it significantly with respect to model-based techniques. For example, our SDL approach needs 83% less time than a state-of-the-art model-based connectivity-maintenance algorithm when managing a team of ten drones.