Runtime Active Safety Risk-Assessment of Highly Autonomous Vehicles for Safe Nominal Behavior 2020-01-0107
Fatal crashes involving automated driving systems, raised the concern of minimum standard requirement for safety, reliability and performance of Autonomous Driving System (ADS)/Advanced Driver Assistance System (ADAS) before this cutting-edge technology take on public roads. However, limited knowledge and understanding of the complex driving scenario followed by highly uncertain behavior of dynamic occupants introduce enormous challenges in design and analyzing the required minimum safety function to insure necessary level of safety. Hence, in order to overcome the challenges to ensure necessary safety requirements of AD/ADAS systems we propose a runtime active safety assurance module to provide required necessary level of safety as well as to overcome the limitation of safety by design risk of an AD/ADAS system. The proposed runtime active assurance safety performs dynamic risk assessment of “Sensing Risk, Planning risk and Action Risk”; to determine whether specified normal performance can be achieved or not in order to guarantee minimal risk maneuver in a given driving scenario. The dynamic risk assessment of AD/ADAS system that comprises of sensing, planning and acting module is based on the operational design domain (ODD) knowledge derived from environment perception sensor capability, environment perception algorithm requirement and capability, dynamic object behavior prediction algorithm requirement and capability and finally smooth and collision free maneuver requirement. So, the main concept behind runtime active safety assurance module is runtime derivation of situational and conditional set of contracts based on the given driving scenario and AD/ADAS ODD; fulfillment or violation of which can help in runtime dynamic risk assessment of automated driving system to plan safe behavior such that necessary safety requirements can be guaranteed. Finally, through experiment we show that proposed runtime active assurance safety module can handle complex driving scenario, and present simulation and experimental results that emphasizes the importance of the proposed runtime safety assurance module and shows that the proposed system is capable of performing runtime dynamic risk assessment in order to keep the automated driving systems always within the safe sate that is the automated driving system always perform within its ODD.
Swarn Singh Rathour, Tasuku Ishigooka, Satoshi Otsuka, RAUL MARTIN