Trajectory Optimization for Multi-UAV-Enabled Wireless-Powered MEC with Graph-Based MADRL

Mobile edge computing (MEC) and wireless power transfer (WPT) are promising techniques for enhancing computational capabilities and extending the operational lifetime of wireless devices in Internet of Things (IoT) networks. In unmanned aerial vehicle (UAV)-enabled wireless-powered MEC (WPMEC) systems, UAVs provide computation offloading and WPT services to ground energy harvesting (EH) IoT devices. This paper investigates cooperative trajectory optimization for multiple UAVs in UAV-enabled WPMEC systems to improve both offloading efficiency and EH performance. UAV agents learn a decentralized trajectory policy using graph-based multi-agent deep reinforcement learning (MADRL), with actions adaptively selected in real time by the proposed algorithm based on local observations. Graph attention networks (GATs) are employed to dynamically weight neighboring node information to improve decision-making in MADRL. Simulation results demonstrate that the proposed algorithm significantly outperforms benchmark schemes in terms of both offloading and EH metrics.