Research Initiation Award: Leveraging Machine Learning for Secure and Privacy-Preserving Vehicular Cyber-Physical Systems (VCPS)

Connected and automated vehicles depend on reliable communication among vehicles, roadway infrastructure, sensors, and control systems. Cyberattacks can block safety messages, alter information exchanged inside a vehicle, or expose sensitive location and driving data. The project will develop machine learning methods to improve the security, resilience, privacy, and safety of connected transportation. Research activities will address three related needs: protecting wireless vehicle communication among vehicles from deliberate interference, detecting malicious activity in the communication bus inside vehicles, and protecting traffic, trajectory, and shared learning data. The research team will evaluate the methods under realistic cyber-attacks and changing driving conditions. Expected outcomes include safer vehicle communication, faster detection of harmful messages, stronger privacy protection, reusable research tools, and new knowledge for secure intelligent transportation. The project will also provide research training for graduate and undergraduate students and will share research findings, educational materials, and data with broader research and education communities. The technical research consists of three coordinated components. The first will develop online learning methods that use inter-vehicle distance to guide channel selection, channel sensing, analysis of adaptive attackers and defenders, and centralized scheduling under several forms of deliberate interference. The second will develop real-time intrusion detection for the Controller Area Network by analyzing both message content and patterns over time, while federated learning will combine locally trained models without transferring raw vehicle data to a central location. The third will generate synthetic traffic data, apply location-sensitive privacy protection to vehicle trajectories, and protect shared learning from malicious model updates and attempts to infer private information. Evaluation will use theoretical analysis, integrated traffic and wireless communication simulations, public and newly collected vehicle data, controlled attack injection, and comparisons with established methods. Quantitative measures will assess learning performance, safe-distance violations, attack-detection accuracy, privacy protection, and resistance to attacks on shared learning. Expected contributions include methods connecting cyber observations with vehicle physical states, practical tools for secure communication and private data sharing, and expanded research and education capacity in machine learning and transportation cybersecurity. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.