Energy-Efficient UAV-Driven Multi-Access Edge Computing: A Distributed Many-Agent Perspective

• Published in: IEEE transactions on communications p. 1
• Main Authors: Li, Yuanjian, Madhukumar, A. S., Ernest, Tan Zheng Hui, Zheng, Gan, Saad, Walid, Hamid Aghvami, A.
• Format: Journal Article
• Language: English
• Published: IEEE 18.03.2025
• Subjects: Autonomous aerial vehicles; Costs; Energy efficiency; energy efficiency maximization; Multi-access edge computing; Multi-access edge computing (MEC); multi-agent deep reinforcement learning (MADRL); Optimization; path planning; Processor scheduling; Resource management; Servers; Training; Trajectory; unmanned aerial vehicle (UAV)
• ISSN: 0090-6778; 1558-0857
• DOI: 10.1109/TCOMM.2025.3552746

In this paper, the problem of energy-efficient unmanned aerial vehicle (UAV)-assisted multi-access task offloading is investigated. In the studied system, several UAVs are deployed as edge servers to cooperatively aid task executions for several energy-limited computation-scarce terrestrial user equipments (UEs). An expected energy efficiency maximization problem is then formulated to jointly optimize UAV trajectories, UE local central processing unit (CPU) clock speeds, UAV-UE associations, time slot slicing, and UE offloading powers. This optimization is subject to practical constraints, including UAV mobility, local computing capabilities, mixed-integer UAV-UE pairing indicators, time slot division, UE transmit power, UAV computational capacities, and information causality. To tackle the multi-dimensional optimization problem under consideration, the duo-staggered perturbed actor-critic with modular networks (DSPAC-MN) solution in a multi-agent deep reinforcement learning (MADRL) setup, is proposed and tailored, after mapping the original problem into a stochastic (Markov) game. Time complexity and communication overhead are analyzed, while convergence performance is discussed. Compared to representative benchmarks, e.g., multi-agent deep deterministic policy gradient (MADDPG) and multi-agent twin-delayed DDPG (MATD3), the proposed DSPAC-MN is validated to be able to achieve the optimal performance of average energy efficiency, while ensuring 100% safe flights.