The clustering activities of users give rise to dynamically changing ground hotspots in both space and time, posing significant challenges for providing communication services to dynamic hotspot areas. UAV swarms, with their ease of deployment and high maneuverability, offer promising solutions. However, UAVs acting as aerial base stations (BSs) are constrained by their limited computing resources and coverage range. Therefore, it is crucial to plan the flight trajectories of UAVs rationally to provide extensive coverage and maximize their resource utilization. We formulate the trajectory planning of multiple UAVs as a multi-objective optimization problem to maximize the ground user coverage, fairness, and UAV resource utilization. We propose a graph attention-enhanced deep reinforcement learning-based algorithm for the trajectory planning of multiple UAVs. The algorithm relies on a centralized training and distributed execution architecture for the flight control of multiple UAVs, enabling UAVs to fully explore the unknown area to learn the distribution of hotspot regions and perceive the optimal altitude to maximize coverage. Additionally, the algorithm incorporates a graph attention encoder to enhance UAV perception of the surrounding environment, addressing the issue of limited observation range. Extensive simulation results demonstrate that our proposed algorithm significantly outperforms baseline algorithms. The proposed algorithm achieves a user coverage rate of 95% and a coverage fairness index close to 1.
Graph Attention Enhanced Multi-UAV Trajectory Planning: A Multi-Agent Deep Reinforcement Learning Algorithm
Lect. Notes Electrical Eng.
International Conference on Autonomous Unmanned Systems ; 2024 ; Shenyang, China September 19, 2024 - September 21, 2024
Proceedings of 4th 2024 International Conference on Autonomous Unmanned Systems (4th ICAUS 2024) ; Kapitel : 10 ; 98-108
28.03.2025
11 pages
Aufsatz/Kapitel (Buch)
Elektronische Ressource
Englisch
Parking task allocation and trajectory planning system based on multi-agent reinforcement learning
Europäisches Patentamt | 2023
|Transportation Research Record | 2025
|