Functional analysis of production grade and certifiable self-driving systems

Abstract:

This paper proposes to decompose an ADAS (Advanced Driver Assistance System) into several layers in order to maximize the reuse of certified and critical components while enabling car manufacturers to still have some
differentiation factors in it. The lower layers abstract the vehicle sensors and actuators so that all the above layers can be reused on any other vehicle. The middle layers focus on understanding the vehicle surrounding and guaranteeing its safety. The upper layers focus on getting the user to its destination and can be customized by the manufacturer in order to create a brand identity. We investigate the functional aspects of such a decomposition.

Furthermore, we provide an extensive analysis of camera sensor requirements for a certification process based on a comparison with human performance. We add some considerations for the use of ASIL (Automotive Safety Integrity Level) and ACAS (Airborne Collision Avoidance System) certification. Finally, during the course of this paper, we make some comments on the certification possibilities of the algorithms involved in an ADAS.

Created by: Sébastien Eskenazi, computer vision expert at FPT Software