Machine Learning-Enhanced Scalable Architectures for Safe and Connected Autonomous Vehicles

Abstract

Autonomous vehicles (AVs) rely on adaptive, data-driven decision-making to advance intelligent transportation systems. While machine learning (ML) enhances perception, prediction, and cooperation in dynamic environments, the large-scale deployment of connected AVs (CAVs) demands scalable computational frameworks that guarantee safety, reliability, and cybersecurity. This paper proposes a novel layered architecture that integrates deep reinforcement learning for adaptive decision-making, federated learning for distributed and privacy-preserving model updates, and graph neural networks (GNNs) for modeling cooperative vehicle interactions. A dedicated safety assurance layer is incorporated to bolster reliability through real-time anomaly detection, uncertainty quantification, and fail-safe redundancy. The system is evaluated using a hybrid SUMO-CARLA simulation framework and real-world datasets (KITTI, Argoverse). Resz`ults demonstrate a 21% increase in decision accuracy, a 34% reduction in latency, and a 50% decrease in collision rates compared to baseline systems. This work provides a comprehensive and scalable framework that significantly enhances the safety and efficiency of connected autonomous vehicle ecosystems