Optimal trajectory of autonomous flying base stations via reinforcement learning