Federated Learning in Vehicular Networks for Resource Management, Privacy, and Security: Challenges and Future Directions

Federated Learning is presented as a transformative and efficient solution to the limitations of centralised machine learning practices in vehicular networks - offering enhanced privacy, efficiency and capturing the essence of distributed intelligence. Communication overhead from frequent model updates, data heterogeneity or non-IID data distributions degrading model convergence, privacy vulnerabilities such as model poisoning or gradient leakage, and high mobility leading to uneven participation of edge devices with inconsideration to system dynamics. To address these concerns areas such as Hierarchical Federated Learning architectures advanced resource management, bandwidth management via gradient sparsification, security enhancements using blockchain supported lightweight consensus and privacy-preserving techniques like differential privacy, homomorphic encryption were explored in detail which in turn gave rise to future research directions to be highlighted. The paper delineates a forward-looking research agenda that includes joint optimisation of resource allocation and addressing tradeoffs between energy, bandwidth, latency, and privacy. It also focuses on the development of adaptive, context-aware orchestration frameworks for dynamic structures, ensuring model interpretability for robustness and safety-critical applications, with particular attention to autonomous driving, traffic management, energy demand optimisation, and privacy concerns. The study finds that federated learning reduces communication overhead, preserves privacy, and supports applications such as autonomous driving, traffic management, and EV energy optimization. It also identifies blockchain, hierarchical FL, and differential privacy as key enablers while highlighting challenges in data heterogeneity, resource constraints, and real-time performance.