A Novel Framework for Modeling and Synthesizing Stealthy Cyberattacks on Driver-Assist Enabled Vehicles

While the first generation of driver-assist enabled vehicles, i.e., adaptive cruise control (ACC) vehicles, are becoming increasingly available, the emerging ACC technologies open a door for malicious cyberattacks, where a select number of ACC vehicles are compromised to drive in an adversarial fashion, degrading the performance of transportation systems. Many prior studies have assumed constant or stochastic attacks without much consideration of their malicious and stealthy nature. Consequently, some attacks may even act favorably to the compromised vehicles, appearing to be an unreasonable practice. To this end, we develop a novel framework for modeling and synthesizing a broad class of potential attacks with practical interpretation considering the attacker perspective. Being able to model and characterize malicious attacks on ACC vehicles is the first step towards developing effective detection and mitigation strategies as ACC vehicles continue to increase in the market. In this study, we first present a general framework describing mixed traffic involving ACC and human-driven vehicles (HDVs) based on car-following dynamics. Under this framework, a class of potential false data injection attacks on ACC sensor measurements are mathematically modeled and incorporated into traffic flow dynamics. Further, we analytically characterize their malicious and stealthy nature, resulting in a class, i.e., ${\mathcal{C}}$, of physically interpretable attacks. To illustrate the modeling mechanism, we conduct numerical experiments to study how attacks drawn from the set ${\mathcal{C}}$ and its complement impact car-following dynamics. In addition, the energy impact of attacks from ${\mathcal{C}}$ on traffic flow is also examined considering different levels of attack severity.