An enhanced MADDPG framework for joint energy and QoS optimization in UAV-assisted vehicular edge computing system

The objective of this study is to address the challenges posed by high energy overhead and the complexity of ensuring quality of service (QoS) for vehicular edge computing in dynamic environments. To this end, this paper investigates the task offloading problem for vehicular edge computing networks in urban areas where there are task offloading hotspots. The objective is twofold: first, to minimize the system energy expenditure, and second, to ensure the service quality. To this end, Unmanned Aerial Vehicles (UAVs) are introduced as mobile offloading nodes to physically reduce the signal transmission distance and lower the system energy consumption. To this end, we propose a resource allocation optimization framework centered on a novel Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm. Our algorithm’s novelty lies in its integration of a dual-critic mechanism for robust and stable Q-value estimation and a maximum entropy framework to enhance exploration efficiency in complex environments. This intelligent algorithm is synergistically coupled with a clustering-based UAV deployment strategy to handle this dynamic problem. This strategy dynamically and autonomously achieves the optimal resource allocation and UAV deployment, with the objective of reducing the system energy overhead and guaranteeing the QoS. Simulation results demonstrate that this framework significantly enhances resource allocation efficiency. Compared to the original MADDPG algorithm, it reduces task costs by 24%, and compared to the fixed offloading position scheme, it reduces task costs by 31.3%. This study offers a valuable reference point and practical insights for reducing energy overhead and optimizing resource allocation in edge computing for vehicular networking.