Cybersecurity of connected and automated vehicles via traffic anomaly detection

Project summary:

Connected and automated vehicles (CAVs) provide new opportunities for malicious actors to compromise vehicle security and compromise traffic flow. While obvious hacks that cause crashes may be easy to identify, other vehicle compromises may be more stealthy and evade detection. Specifically, attacks that compromise inter-vehicle communication such vehicle-to-vehicle (V2V) or vehicle-to-infrastructure (V2I) messages may introduce increase crash risk and reduce traffic safety. We propose to study such attacks to CAV communication networks, and develop anomaly detection tools that can identify if an attack is present, based on vehicle trajectories and messages sent across the communication network (e.g., 5G connected vehicles). Specifically, we propose to use microsimulation to simulate traffic flow both of typical mixed autonomy traffic as well as traffic where some of the automated vehicles have been compromised and are sending compromised communications to other vehicles. The communication layer will also be modeled independently, with vehicles sharing basic safety messages (BSMs) across the network. Potential cyberattacks will be implemented in simulation, where compromised messages are communicated across the network, and the resulting traffic and communication data as well as traffic and communication data from uncompromised traffic flow will be compared to understand the potential impact of such attaches. Furthermore, the generated synthetic data will be used to develop anomaly detection techniques that leverage advancements in neural networks and autoencoders to identify atypical traffic and communication data.