Enhancing UAV-assisted vehicle edge computing networks through a digital twin-driven task offloading framework

Abstract

Enhancing the task offload performance of UAV-assisted Vehicular Edge Computing Networks (VECNs) is complex, especially in vehicle-to-everything (V2X) applications. These networks rely on UAVs and roadside units (RSUs) to offload heavy computational tasks and reduce the load on the on-board systems. However, UAV-assisted VECNs face severe challenges from heterogeneous offload node resources and dynamic edge network environments in providing low-latency and high-response task offloading, especially during traffic congestion or infrastructure failures. In this paper, we propose a digital twin (DT)-driven task offloading framework for UAV-assisted VECNs. The aim of the proposed framework is to improve the global performance of VECN task offloading under limited computational and communication resource constraints. Firstly, we construct a decentralized offloading decision-centralized evaluation task offloading framework for UAV-assisted VECNs based on the asynchronous advantage actor-critic (A3C) algorithm. Secondly, we integrate the graph attention networks (GAT) into the framework to incorporate the dynamically changing DT network topology information into the state evaluation of VECNs. By simulating a DT-driven multi-UAV cooperative system and comprehensive evaluation of real-world task request datasets. The framework has a better task throughput rate and stability when performing task offloading in local resource overload and dynamic edge environment scenarios.