Adaptive Resilient Control for Autonomous Vehicles Steering System against False Data Injection Attacks

Autonomous vehicle (AV) whose steering system is subjected to false data injection (FDI) attacks will quickly lose stability and deviate from the correct trajectory. This paper presents an adaptive resilient control (ARC) method that integrates a learning-based stochastic model predictive control (SMPC) to mitigate the impact of FDI attacks on the AV steering system. First, a nominal error model is introduced to describe the lateral tracking trajectory control of AV. Second, a real-time online learning strategy is devised to continuously update the vehicle dynamics. Gaussian process (GP) is utilized to detect unmodeled deviations resulting from FDI attacks and incorporate the training outcomes into the nominal error model, thereby obtaining a more accurate estimated model. Then, the estimated model is integrated into SMPC to optimize motion control for trajectory tracking. Finally, simulation tests are conducted using the CarSim to confirm the effectiveness of the proposed method.