Adaptive Task Offloading in Vehicular Edge Computing Networks Based on Deep Reinforcement Learning

Abstract

To support the delay-sensitive application for vehicular communications, edge servers are adopted and deployed in the roadside units to provide abundant computing resources for vehicles forming vehicular edge computing networks (VECNs). This paper tries to optimize the task offloading scheme to reduce the total task processing delay as much as possible in VECNs. Considering the dynamic nature of vehicular networks, we assume a continuous-time scenario of VECNs and propose an adaptive task offloading policy. Compared with the existing works that design the task offloading policy per single moment, the proposed policy can guarantee a minimum long-term processing delay. To minimize the task processing delay, we formulate an optimization problem and propose a deep reinforcement learning based method to find out the optimal offloading strategy. Simulation results demonstrate that the proposed method outperforms all baseline counterpart algorithms.