Asynchronous Federated Learning Technique for Latency Reduction in STAR-RIS Enabled VRCS

With the advent of smart and autonomous vehicles, a number of novel data-intensive and latency-critical vehicular communication applications have emerged. However, dynamic vehicular mobility and urban environments introduce severe propagation challenges, leading to increased latency. In order to reduce latency in Vehicle Road Cooperative Systems (VRCS), this research introduces a unique architecture that combines Asynchronous Federated Learning (AFL) with Simultaneously Transmitting and Reflecting Reconfigurable Intelligent Surfaces (STAR-RIS). The proposed system leverages a Markov Decision Process (MDP)-based optimization framework to minimize latency by jointly optimizing STAR-RIS elements and offloading decisions. Our approach allows vehicles to asynchronously update global models, ensuring robust learning while adapting to dynamic network conditions. The simulation results show that the recommended strategy provides at least a 20 % reduction in latency in AFL when compared to FL.