Joint Optimization of Sensing and Data Offloading in Digital Twin-Assisted Internet of Vehicles

Digital Twin (DT) technology can build consistent virtual replicas of physical infrastructures by collecting sensed environmental information, and further offers a powerful paradigm for real-time monitoring, simulation, and optimization of vehicular networks. However, there exists a natural trade-off between information sensing and DT reconstruction. Specifically, insufficient sensing or unreliable sensing results may increase DT inconsistency, while exhaustive sensing can incur substantial energy overhead, especially for fast-changing, resource-constrained, and dynamic Internet of Vehicles (IoV) systems. To address this challenge, we investigate a multi-layer DT-IoV framework based on cloud-edge-vehicle collaborations. Through the joint design of task sensing, vehicle selection, and data offloading strategies, we maximize the consistency of DT reconstruction under limited resources. Furthermore, we develop a spatio-temporal similarity-based hierarchical clustering reinforcement learning (HC-RL) algorithm in a mesh-free environment to minimize the selection of invalid vehicles. Simulation results demonstrate the effectiveness of our proposed framework and method in preserving DT model fidelity and minimizing energy consumption and latency.