Decentralized Multi-UAV Management in Mobile Edge Computing: A Hierarchical Reinforcement Learning Approach

The integration of drones in Mobile Edge Computing (MEC) has opened new possibilities for enhancing network performance in various fields, such as agriculture, environmental monitoring, and disaster management. However, effective coordination of multiple Unmanned Aerial Vehicles (UAVs) in dynamic and distributed environments poses significant challenges, especially in mobility management and resource allocation. To address these issues, we propose a Decentralized Hierarchical Policy Framework inspired by hierarchical reinforcement learning. This framework employs a two-level structure, consisting of high-level communication coordinators and low-level executors, allowing UAVs to make decentralized decisions based on local observations and partial communication. Our approach models the multi-UAV MEC problem as a Decentralized Partially Observable Markov Decision Process with communication, enabling efficient task allocation and resource management. Experimental evaluations demonstrate that the proposed method improves MEC service coverage, reduces latency, and optimizes UAV energy consumption. Compared to state-of-the-art Multi-Agent Reinforcement Learning methods, our approach achieves superior performance in dynamic and large-scale environments.