Diversification of Autonomous Vehicle Driving Behavior

Driver behavioral analysis can be described as the process of analyzing sensor data collected from a vehicle to identify unique facets of the manner the vehicle handles while under the control of a driver. This analysis can be used to either uniquely identify a driver or differentiate him/her from other drivers. Driver behavioral analysis has practical applications in insurance, safety and security. Autonomous vehicle controllers are software systems that are static by nature. This implies that while driver behavior analysis can be carried out on autonomous vehicles, the unique driving patterns that can be used to differentiate human drivers will not be possible with autonomous vehicles controlled by the same agent. This scenario is undesirable from a security viewpoint. It is widely accepted that the predictability of software systems makes them vulnerable, this is more so if a large of number of users utilize the software system. Based on the above, in this paper, we argue for the necessity of creating diversity in the driving behavior of autonomous vehicles. We present a conceptual framework that lays out how this can be accomplished at low cost, with high flexibility and without compromising one of the major advantages of autonomous vehicles - safety. Using the example of a remote vehicle hijack, an attack scenario that has shown to be practical in recent times, we highlight an application of our proposed model. We show how having diverse behavior is advantageous in creating a fail-safe remote vehicle hijack detection system.