SOTIF-Oriented Perception Evaluation Method for Forward Obstacle Detection of Autonomous Vehicles

Abstract

The inclusion of autonomous vehicles into our society requires not only mature automated technologies but also adequate confidence in safety guarantees. Safety of the intended functionality (SOTIF) is an incipient important field that focuses on risk led by performance limitations rather than faults. There are still plenty of gaps in addressing it, especially when concerning perception ability. This article introduces a method to evaluate SOTIF-oriented perception effectiveness for forward obstacle detection of autonomous vehicles. Both inevitable existence uncertainty and state uncertainty in perception results are taken into account, and a formal SOTIF requirement specification is defined. Based on the safety distance model, we further deduce the minimum required perception area and the perception error requirements under a given driving situation to refine the SOTIF requirement specification. Then, a SOTIF-oriented unified sensing model is established by combining sensing space and ability characteristics. The system-level perception ability of an autonomous vehicle is subsequently deduced with the assumption that every available sensor is described in the form of our proposed sensing model. Evaluation can be executed according to the above methods that support determining the SOTIF requirements and the system-level perception ability, and the entire evaluation process is presented as an algorithm in this article. Finally, a practical case generated according to a well-known autonomous vehicle prototype is used to prove the effectiveness of the method. Evaluation results can serve as suggestions to improve the overall SOTIF level for perception.