A Reinforcement Learning-Based Cooperative Adaptive Cruise Control Under False Data Injection Attacks

Abstract

Connected and automated vehicles (CAVs) are introduced to enhance the safety, efficiency, and mobility of transportation systems by leveraging advanced onboard sensors and wireless communication technologies. Cooperative adaptive cruise control (CACC), a fundamental application of CAVs, enhances adaptive cruise control (ACC) by treating multiple vehicles in close proximity as a platoon. Each vehicle within the platoon is equipped with a controller that leverages sensor-based measurements and vehicle-to-vehicle (V2V) communication to maintain safe and optimal inter-vehicle spacing. Although this system has significant potential to enhance traffic safety and efficiency, it remains vulnerable to false data injection (FDI) attacks, which can compromise system safety. Since the FDI attack is unknown and nonlinear, the CACC should be able to estimate and mitigate it in real-time. There exist several techniques in the literature to detect and estimate FDI attacks in real-time. However, none of them provides an asymptotic coverage. To address this issue, this paper presents a novel adaptive critic control and FDI attack estimation design leveraging reinforcement learning and Lyapunov stability analysis to mitigate the adverse effects of FDI attacks and bounded disturbance. This method yields semi-global asymptotic stability results. Theoretical analysis and simulation results are provided to illustrate the performance of the developed controller.