Fuzzing Malicious Driving Behavior to find Vulnerabilities in Collision Avoidance Systems

In security research, it is difficult to design a system that can anticipate future actions by attackers. This is particularly true for cyber-physical systems because adversaries can create an infinite variety of physical environmental conditions to attack a system. We need to develop a more principled approach for evaluating and testing control algorithms against physical adversarial situations to design attack-resilient control systems. In this paper, we leverage the tools created by the falsification community and re-purpose them to consider finding attacks. Falsification provides a formal methodology to identify inputs that will violate the requirements of a system. In particular, we look at how cruise control algorithms respond to actions taken by malicious drivers that want to create an accident (while avoiding crashing themselves). We propose a scenario-based simulation and metric temporal logic to identify vulnerabilities in three cruise control with collision avoidance control strategies.