Formal Safety and Cyber-Security in Mixed-Autonomy Traffic

Abstract

This work studies formal safety and cyber-security in mixed-autonomy traffic flows. Mixed-autonomy traffic flows have a mix of human drivers and automated vehicles (AVs). This work considered three emergent problems in this context. 1) Are adaptive cruise control (ACC) vehicles string-stable? Through experimentation on commercially available ACCs we found them to not be string stable. 2) Are ACCs stealthy vectors of attack against traffic? Through building a high fidelity simulation testing environment we found that ACCs may be used to attack traffic flows in a stealthy manner. 3) Can control barrier functions (CBFs) guarantee the safety of AVs in live traffic? Through experimentation we found that CBFs do not provide formal safety in real world contexts, but can practically be used to supervise otherwise unsafe AV algorithms.