Adaptive Digital Twin for Vehicular Edge Computing and Networks

To better support the emerging vehicular applications and multimedia services, vehicular edge computing (VEC) provides computing and caching services in proximity to vehicles, by reducing network transmission latency and alleviating network congestion. However, current VEC networks may face some implementation challenges, such as high mobility of vehicles, dynamic vehicular environment, and complex network scheduling. Digital twin, as an emerging technology, can make the virtual representation of physical networks to predict, estimate, and analyze the real-time network state. In this paper, we integrate digital twin into VEC networks to adaptively make network management and policy schedule. We first introduce the framework of VEC networks and present the key problems in a VEC network. Next, we give the concept of digital twin and propose an adaptive digital twin-enabled VEC network. In the proposed network, digital twin can enable adaptive network management via the two-closed loops between physical VEC networks and digital twins. Further, we propose a digital twin empowered VEC offloading problem with vehicle digital models and road side unit (RSU) digital models. A deep reinforcement learning (DRL)-based offloading scheme is designed to minimize the total offloading latency. Numerical results demonstrate the effectiveness of the proposed DRL-based algorithm for VEC offloading.