Privacy-Preserving Federated Reinforcement Learning for Autonomous Connected Vehicles

Abstract

Autonomous connected vehicles continuously exchange large volumes of sensor and traffic information to improve navigation and driving safety. However, centralized learning approaches expose sensitive vehicular data to privacy risks and communication bottlenecks. This research proposes a privacy-preserving federated reinforcement learning framework that enables autonomous vehicles to collaboratively learn optimal driving policies without sharing raw data. Differential privacy and secure aggregation mechanisms are integrated into the federated learning process to safeguard user information while maintaining model performance. Deep Reinforcement Learning algorithms optimize route planning, traffic management, and collision avoidance using decentralized knowledge sharing. Simulation results demonstrate enhanced driving efficiency, reduced communication overhead, and improved privacy protection compared to centralized autonomous driving systems. The proposed framework contributes to secure and intelligent transportation infrastructures supporting future smart mobility.