DRIVE: Dynamic Resource-Aware Intelligent Task Offloading in Vehicular Edge Computing

Abstract

With the swift evolution of intelligent transportation systems, Vehicular Edge Computing (VEC) has developed rapidly. In VEC environments, the lack of dynamic awareness of spatiotemporal-varying heterogeneous resources results in high task offloading latency in high mobility scenarios. Therefore, this paper proposes an intelligent task offloading method based on dynamic resource awareness in vehicular edge computing environments by using Twin Delayed Deep Deterministic Policy Gradient (TD3). The proposed method first establishes a cloud-edge-end multi-dimensional collaborative framework to support the perception and coordination of unevenly distributed multi-dimensional dynamic resources throughout task offloading. Secondly, a multi-objective model is established with the core objective of minimizing task execution latency, while simultaneously considering computational and bandwidth resources as well as link timeliness constraints. Finally, the TD3 algorithm, which effectively mitigates Q-value overestimation bias through mechanisms such as clipped double Q-learning, is employed to derive efficient and rational task offloading decisions. Experimental results on datasets from Beijing and Shanghai telecom base stations and taxi trajectory, show that the proposed method achieves lower weighted task latency than existing baseline approaches in the field.