Traffic models with adversarial vehicle behaviour

Abstract: We examine the impact of adversarial actions on vehicles in traffic. Current advances in assisted/autonomous driving technologies are supposed to reduce the number of casualties, but this seems to be desired despite the recently proved insecurity of in-vehicle communication buses or components. Fortunately to some extent, while compromised cars have become a reality, the numerous attacks reported so far on in-vehicle electronics are exclusively concerned with impairments of a single target. In this work we put adversarial behavior under a more complex scenario where driving decisions deluded by corrupted electronics can affect more than one vehicle. Particularly, we focus our attention on chain collisions involving multiple vehicles that can be amplified by simple adversarial interventions, e.g., delaying taillights or falsifying speedometer readings. We provide metrics for assessing adversarial impact and consider safety margins against adversarial actions. Moreover, we discuss intelligent adversarial behaviour by which the creation of rogue platoons is possible and speed manipulations become stealthy to human drivers. We emphasize that our work does not try to show the mere fact that imprudent speeds and headways lead to chain-collisions, but points out that an adversary may favour such scenarios (eventually keeping his actions stealthy for human drivers) and further asks for quantifying the impact of adversarial activity or whether existing traffic regulations are prepared for such situations.