Communication-Efficient Federated Learning Framework for Autonomous Vehicles

As autonomous vehicles increasingly rely on real-time perception and decision-making, the demand for decentralized and efficient machine learning has intensified. Federated Learning (FL) offers a promising paradigm by enabling collaborative model training across distributed edge nodes, such as vehicles, without exposing raw data. However, deploying FL in autonomous driving environments remains challenging due to the substantial communication overhead between edge devices and coordinating servers, particularly under dynamic mobility and variable-bandwidth conditions. This paper presents a communication-efficient FL framework tailored for edge-based autonomous vehicle networks. The proposed approach reduces uplink and downlink costs through optimized parameter exchange and adaptive update strategies while maintaining model accuracy. Experimental results demonstrate improved scalability and communication efficiency, highlighting the potential of FL for intelligent, connected, and resource-constrained autonomous mobility systems under non-independent and identically distributed (non-IID) data distribution.