Trajectory Design for UAV Communications with No-Fly Zones by Deep Reinforcement Learning

This paper studies the trajectory design problem for the cellular-connected unmanned aerial vehicle (UAV) with limited energy, which aims at maximizing the uplink transmission rate from multiple ground users in urban environments with no-fly zones (NFZs). We first argue that the successive convex approximation-based (SCA-based) conventional trajectory design method via formulating and solving optimization problems face challenges, and then we formulate the trajectory design problem for rate maximization as a Markov Decision Process and propose a deep reinforcement learning-based (DRL-based) solution. Simulation results show that the proposed DRL has similar performance to the SCA-based conventional method with regularly shaped NFZ constraints. Moreover, simulation results in a scenario with an irregular NFZ show that the designed trajectories of the proposed DRL can effectively serve users and detour the NFZ.