Ground Power for Sky Computing: Sustaining UAV MEC With Mobile UGV Charging

Abstract

Computation-intensive and latency-sensitive applications often exceed the processing capabilities of User Devices (UDs). Unmanned Aerial Vehicles (UAVs) can assist Mobile Edge Computing (MEC) by enabling task offloading, thereby reducing service latency. However, the limited onboard energy of UAVs restricts continuous operational endurance. This limitation is exacerbated in remote or dynamic environments where fixed charging infrastructure is impractical. To overcome these challenges, this paper introduces a novel Unmanned Ground Vehicle (UGV)-based mobile charging scheme for UAV-enabled MEC system. In this scheme, UGVs function as mobile charging stations to power UAVs, thereby extending their operational duration and ensuring service continuity. Specially, a Trajectory-Prediction Enhanced Hierarchical Deep Reinforcement Learning (TP-HDRL) framework is developed to jointly optimize task offloading, resource allocation, trajectory planning of UAVs, and charging scheduling of UGVs. By proactively capturing the spatio-temporal features from UD historical trajectories, the framework gains predictive awareness of UD mobility. Additionally, a residual reinforcement learning mechanism is incorporated to accelerate the convergence in high-dimensional action spaces and boost overall performance. Simulation results demonstrate that the proposed solution significantly reduces task completion delay for UDs and improves UAV energy efficiency compared to existing baseline methods.