Zero-Touch MEC Resources for Connected Autonomous Vehicles Managed by Federated Learning

Abstract

This paper presents a Ph.D. thesis proposal for a novel solution in optimizing the placement of Connected Autonomous Vehicles (CAVs) Virtual Network Functions (VNFs) requests in Edge Computing (EC) resources. Our Federated Deep Reinforcement Learning (FDRL) proposal will be designed to improve computation efficiency while minimizing service rejections and maximizing resource utilization, and ensuring the least costly path for CAVs. This approach will also be privacy-preserving, ensuring sensitive data remains secure and enables reliable, low-latency communication between CAVs, EC nodes, and the federated server. By utilizing distributed learning capabilities, FDRL allows multiple vehicles to learn from their local experience and make collective decisions, improving network systems performance.