Detecting Subtle Cyberattacks on Adaptive Cruise Control Vehicles: A Machine Learning Approach

Abstract

With the emergence of vehicles featuring advanced driver-assistance systems like adaptive cruise control (ACC) and additional automated driving functionalities, there has arisen a heightened potential for cyberattacks targeting these automated vehicles (AVs). While overt attacks that lead to collisions are more conspicuous, subtle attacks that slightly modify driving behaviors can cause widespread impacts, including increased congestion, fuel consumption, and crash risks without being easily detected. To address the detection of such attacks, we first present a traffic modeling framework for three types of potential cyberattacks: malicious manipulation of vehicle control commands, data poison attacks, and denial-of-service (DoS) attacks. Subsequently, we examine the consequences of these attacks on both singular vehicle dynamics (micro) and broader traffic flow patterns (macro). We introduce a new anomaly detection model based on generative adversarial networks (GAN) designed for the real-time pinpointing of such attacks using vehicle trajectory data. Numerical results are presented to show the effectiveness of our machine learning strategy in identifying cyberattacks on vehicles equipped with ACC. The proposed approach is observed to outperform contemporary neural network models in detecting irregular driving patterns of ACC vehicles.

Keywords

• Adaptive cruise control (ACC) vehicle
• automated vehicle
• transportation cybersecurity
• machine learning