Deep Reinforcement Learning Based UAV Trajectory Design for Data Collection Scenario with No-Fly Zones

Abstract

Recently, unmanned aerial vehicle (UAV)-assisted communication system has been introduced as a promising paradigm for the future space-aerial-terrestrial integrated communications. In this paper, we investigate an UAV communication system, where the UAV is employed to assist multiple ground loT devices for data collection in the area of interest with the existence of no-fly zones. Unlike existing approaches focusing only on simplified line-of-sigh (LoS)-dominant channel model, this paper considers a more practical probability LoS channel model, which considers path loss and shadowing. On the premise of satisfying the data throughput requirements of all ground loT devices, we intend to minimize the total task completion time by jointly optimizing UAV's trajectory and communication scheduling. To tackle the non-convex and difficult intractable problem, we first transform the original problem into an Markov decision process (MDP) problem, and then we propose a trajectory design solution based on deep reinforcement learning (DRL) algorithm for completion time minimization. The UAV serves as an agent in the process of execution algorithm, interacting with the environment and constantly improving its own mobile strategy. Finally, numerical results demonstrate that the proposed design contributes to significant performance enhancement and can be applied to practical scenarios with no-fly zones.