A Digital-Twin-Based Traffic Guidance Scheme for Autonomous Driving

Burdened by persistent traffic congestion, urban transportation is in a pressing need of more effective traffic guidance schemes. Existing traffic guidance approaches fall short in optimizing benefits, primarily due to their exclusive reliance on current road conditions for decision making and the prevalence of driving egoism within traditional patterns. Autonomous vehicles (AVs), liberated from human control and enhanced by the Internet of Vehicles and edge computing, provide new possibilities for traffic guidance. Nevertheless, it is tough to precisely determine the pertinent information to convey and establish an effective cooperative guidance mechanism in the face of the substantial number of AVs. This article proposes a social value orientation (SVO)-based cooperation mechanism for AVs, through which the driving routes are jointly determined by individual driving demands, local road network conditions, and global benefits. We design a digital twin-based Edge-to-Cloud traffic guidance architecture, leveraging real-time AV decisions and micro-driving characteristics for forthcoming road condition estimation. The hierarchical Edge-to-Cloud structure efficiently mitigates communication and computation overheads in traffic guidance by distributing tasks across different regions. Finally, an innovative method based on inverse reinforcement learning is proposed to address the challenge of adapting guidance policies in response to varying traffic densities and distributions. The simulation results show a 59.1% improvement in the travel achievement ratio under heavy road traffic load, with no significant change in the detour ratio. It indicates an enhanced system driving efficiency, while still safeguarding the individual benefits of AVs.