Connected and automated vehicles (CAVs) promise safer, more efficient roads by combining onboard sensors with wireless communication. A core feature, Cooperative Adaptive Cruise Control (CACC), allows vehicles in a platoon to automatically adjust their speed and spacing. However, these systems are highly vulnerable to false data injection (FDI) attacks, which can disrupt coordination, trigger unsafe distances, and increase collision risk.
This innovation introduces a secure, learning-based control strategy that defends CACC systems against such attacks. The system integrates a nonlinear actor-critic reinforcement learning controller, a neural network-based FDI attack estimator, and Lyapunov stability analysis to achieve asymptotic tracking of vehicle safe distances. Unlike previous solutions that offer only limited stability, this method ensures long-term, adaptive safety for real-world deployment in CAVs. It offers a robust software controller easily integrated into vehicle systems, ready for automotive, fleet, and mobility applications.
This image shows a diagram of the developed adaptive and secure actor-critic cooperative adaptive cruise control (CACC) under false data injection (FDI) attacks, where malicious signals disrupt the received control signal from the lead vehicle. The technology defends against these attacks to maintain stability and ensure safe vehicle coordination.