AoI- and Energy-Efficient Multi-UAV Coordination via Hierarchical Deep Reinforcement Learning

초록

Uncrewed aerial vehicles (UAVs) are increasingly used in Internet of Things (IoT) applications that require timely information acquisition, such as environmental surveillance, wildfire detection, and emergency response. Although multi-UAV systems provide improved coverage and flexibility, efficiently coordinating multiple UAVs to maintain information freshness, measured by the age of information (AoI), while reducing energy consumption remains challenging, especially in large-scale deployments. To address this problem, we propose GALA, a hierarchical deep reinforcement learning (DRL) framework for AoI- and energy-efficient multi-UAV coordination. GALA follows a coarse-to-fine paradigm with a high-level global allocator (GA) for UAV-device assignment and a low-level local navigator (LA) for allocation-conditioned trajectory execution. The GA uses adaptive feature aggregation over a graph-based representation to improve coordination efficiency and scalability, while the LA adopts an AoI-aware decision strategy for effective local control. A two-stage DRL training scheme further improves learning efficiency. Extensive comparisons with state-of-the-art baselines and ablation studies demonstrate the superiority of GALA across diverse settings. Real-world experiments further validate their practical applicability.

키워드