Optimal Trajectory Learning for UAV-Mounted Mobile Base Stations using RL and Greedy Algorithms

This paper designs Artificial Intelligence (AI) method, to determine an optimal trajectory for an Unmanned Aerial Vehicle (UAV) mounted mobile base station to maximize its coverage of distinct users. Determining such an optimal trajectory for arbitrarily distributed users over an area is, in general, difficult and there is no closed-form solution. Since the users are arbitrarily located the method must adapt accordingly. To accomplish this, the problem is formulated in a way that is compatible with Reinforcement Learning (RL) and two AI approaches are designed to learn an optimal trajectory. The first uses Deep Reinforcement Learning (DRL) implemented with a Deep Q-Network (DQN) while the second is a reward-based greedy algorithm. It is shown that these new algorithms significantly outperform the state-of-the-art previously proposed deep learning based approaches. Moreover, the simple AI-based greedy approach is shown to perform close to the DQN-aided DRL algorithm at a much lower computational complexity at least in the type of scenarios considered in this paper.