A Digital Twins-Assisted Multi-Autonomous Vehicle Distributed Collaborative Path Planning Algorithm With Fidelity Guarantee

Autonomous driving is state-of-the-art technology in the field of Internet of Vehicles (IoVs), considered a revolutionary solution to enhance driving safety and comfort. Collaborative path planning for autonomous driving vehicles has not been practically applied due to the low effectiveness of models trained by traditional machine methods. The Digital Twins (DTs) map the state of the autonomous vehicles (AVs) in the real world to the virtual space in order to analyze the vehicle behavior and optimize vehicle decisions. During the DTs synchronization process, the time-varying nature of the physical environment may result in loss of DTs synchronization data. In order to better ensure the fidelity of DTs and the safety of autonomous driving, we propose a DTs-assisted distributed federated reinforcement training framework with fidelity guarantee. In this framework, the DTs leverage its predictive function to fill the lost data during the vehicle synchronization to its twins, ensuring a high fidelity in mapping the vehicle state. Then the model undergoes training through distributed federated reinforcement learning within the DTs environment. The simulation results indicate that our proposed solution not only ensures high fidelity in modeling the digital twins but also enhances the utilization efficiency of vehicle speeds. Additionally, it reduces the collision probability and average task completion time for the vehicle swarm.