Ubiquitous Control Over Heterogeneous Vehicles: A Digital Twin Empowered Edge AI Approach

The forthcoming of automated driving has led to vehicular heterogeneity, where vehicles with different automation levels, including connected and automated vehicles (CAVs), connected vehicles (CVs), and human-driven vehicles (HVs), coexist in the road traffic. However, the design of current traffic control systems fail to account for the vehicular traffic heterogeneity. In addition, the emerging artificial intelligence (AI) based traffic control strategies require large quantities of computation resources and generate high delay, which cannot meet the fault-intolerance requirement of the traffic control system. Therefore, there is an urgent need for constructing a new traffic control framework to jointly satisfy the fault-intolerance requirement and realize ubiquitous control over the heterogeneous vehicles. Inspired by this, a digital twin (DT) empowered edge AI framework is proposed in this article. The DT's virtualization capability is utilized to enable virtual risk assessment and performance analysis under the fault-intolerant traffic control system. In addition, the DT's offline learning capability helps the edge AI understand the road traffic data more intelligently to perform enforced optimizations and decisions. Then, a three-layer vehicle control paradigm is discussed under the proposed framework. In the first layer, deep reinforcement learning (DRL) is utilized to intelligently control the CAV acceleration and lane-changing. In the second layer, indirect CV/HV control can be performed by adjusting the target speed and penetration rate of the DRL-controlled CAVs which are mixed in the traffic flow. In the third layer, vehicle-to-everything (V2X) communications and variable message signs are utilized to realize the direct control of CVs and HVs. Experimental results validate the effectiveness of our proposed control paradigm.