Cyber-Resilient Autonomous Vehicle Control Thwarts Actuator Attacks with Data-Driven Framework

Autonomous vehicles (AVs) are revolutionizing transportation with promises of enhanced safety and efficiency, but their reliance on networked control systems exposes them to insidious cyber threats like false data injection (FDI) attacks. These attacks can manipulate actuator signals, leading to dangerous instabilities in critical functions such as lane-keeping and path-following. A groundbreaking study introduces a cyber-resilient secure control framework that integrates data-driven modeling, event-triggered communication, and fractional-order sliding mode control to actively neutralize these threats. By moving beyond traditional detection s, this approach ensures real-time compensation for adversarial interventions, making AVs more robust in safety-critical scenarios where milliseconds matter.

The proposed ology leverages dynamic mode decomposition (DMD) to extract lateral dynamics directly from real-world sensor data, eliminating the need for error-prone mechanistic models. Experimental data from HORIBA MIRA's Autonomous Vehicle Development Centre, including RTK GPS and LiDAR measurements, were used to construct state matrices via singular value decomposition, ensuring high fidelity in modeling vehicle behavior. An event-triggered transmission scheme reduces communication overhead by transmitting control updates only when predefined error thresholds are exceeded, optimizing bandwidth usage without compromising stability. Additionally, an extended state observer (ESO) estimates actuator attack signals in real-time, enabling the control system to dynamically counteract FDI attacks through a combination of equivalent and switching controllers based on fractional-order sliding mode principles.

Simulation under three scenarios—nominal conditions, attacks without mitigation, and attacks with the proposed secure control—demonstrate the framework's efficacy. In attack scenarios without mitigation, lateral errors surged to nearly 1 meter and control inputs became erratic, highlighting system vulnerabilities. However, with the secure controller, lateral errors remained within ±0.1 meters, and the ESO achieved 92% estimation accuracy, swiftly neutralizing malicious inputs. The event-triggered mechanism maintained communication efficiency, with transmission intervals averaging 0.134 seconds in secured cases compared to 0.080 seconds under unmitigated attacks, showcasing a balance between resource conservation and performance.

Of this research are profound for the automotive and cybersecurity industries, as it addresses the growing concern of model-aware cyber-attacks that evade conventional defenses. By integrating data-driven approaches with secure control, the framework enhances resilience against stealthy FDI threats, potentially reducing accident risks and enabling wider AV adoption. Theoretical analyses using Lyapunov s and linear matrix inequalities guarantee exponential error convergence, providing a solid foundation for real-world implementations in connected and autonomous vehicles, where reliability is paramount.

Despite its strengths, the study acknowledges limitations, such as the assumption of bounded attack magnitudes and potential performance degradation if delays exceed theoretical bounds. Future work could explore adaptive data-driven techniques and online modeling to handle time-varying dynamics and environmental uncertainties. This research, detailed in the source below, marks a significant step toward safer autonomous systems, blending cutting-edge control theory with practical cybersecurity solutions.