Quantifying integrated safety risk in highly automated vehicles: A probabilistic approach to perception sensor uncertainty

Abstract

Highly automated vehicles (AVs) rely on sensor data for target tracking and maintaining a safe separation distance during safety-critical operations such as forward collision avoidance. However, the inherent uncertainty in perception sensor measurements can lead to inaccurate tracking, which poses challenges for ensuring passenger safety. Method: This study proposes a method to quantify integrated safety risk using a probabilistic approach, incorporating various scenarios of perception sensor uncertainty and linking them to corresponding collision risks and subsequent serious passenger injury risks. A novel approach in this method involves subdividing the risk for each distance to the target, as defined by a perception uncertainty model. These risks are then integrated to compute the total safety of the ego-vehicle’s passengers. Results and conclusions: By considering both target presence and absence hypotheses, the algorithm innovatively addresses risks posed by potentially undetected targets, significantly enhancing user protection and advancing AV safety. Practical Applications: The developed algorithm contributes to the integrated safety of AVs by offering guidance on regulating a minimum separation distance or maximum vehicle speed for a given vehicle sensor set, or specifying the sensor specifications that should be equipped on a vehicle. This approach aims to enhance the reliability and safety of automated driving systems, ensuring a higher standard of passenger safety and fostering trust in automated vehicle technologies.