Strategies to Define Reasonable Acceptance Criteria and Validation Targets for SOTIF Assurance

Abstract

<div class="section abstract"><div class="htmlview paragraph">Acceptance criteria and validation target are the most important metrics used to measure/assure safety of the intended function (SOTIF) of an autonomous vehicle or advanced driver assistance system (ADAS). Often acceptance criteria are defined as acceptable number of fatalities, injuries or property damage events in certain hours of operation or for certain mileage driven. Validation target on the other hand is the amount of effort required in terms of hours of operation or mileage to be driven to show that the acceptance criteria is met. Although existing research details about potential values for acceptance criteria and validation target, they overlook various factors such as operational design domain, operational lifetime of a vehicle, average mileage of a vehicle, and length of roads. As a result, often acceptance criteria values are very small (e.g., 10^-12 incidents/h or mi) and validation targets are very large (e.g., 10¹³ miles). However, these targets may not be practical to distribute within various testing avenues/platforms and provide as evidence for a given version. In this paper, we discuss the various factors and strategies we can consider for defining reasonable and practically achievable acceptance criteria and validation targets. For example, the number of miles driven by motor vehicles in United States in 2020 and the number of accidents occurred in those miles is given based on distance travelled by around 276 million vehicles. Will the distance and number of accidents remain the same even as the fleet size is 100 vehicles? In this paper, we aim to resolve such questions. Using an illustrative example, we demonstrate the usefulness of our proposed factors and strategies.</div></div>