A Method for Determining Mileage Accumulation for Robustness Validation of Advanced Driver Assistance Systems (ADAS) Features

Abstract

<div class="section abstract"><div class="htmlview paragraph">Robustness testing of Advanced Driver Assistance Systems (ADAS) features is a crucial step in ensuring the safety and reliability of these systems. ADAS features include technologies like adaptive cruise control, lateral and longitudinal controls, automatic emergency braking, and more. These systems rely on various sensors, cameras, radar, lidar, and software algorithms to function effectively. Robustness testing aims to identify potential vulnerabilities and weaknesses in these systems under different conditions, ensuring they can handle unexpected scenarios and maintain their performance.</div><div class="htmlview paragraph">Mileage accumulation is one of the validation methods for achieving robustness. It involves subjecting the systems to a wide variety of real-world driving conditions and driving scenarios to ensure the reliability, safety, and effectiveness of the ADAS features. Following ISO 21448 (Safety of the intended functionality-SOTIF), known hazardous scenarios can be tested and validated through robustness testing and validation. Unknown hazardous scenarios can be exposed and identified as known hazardous scenarios through accumulated miles. However, determining the mileage needed for acceptance still poses a challenge.</div><div class="htmlview paragraph">This paper presents a potential methodology utilizing the Sequential Probability Ratio Test (SPRT) as acceptance criteria to determine the required mileage accumulation and to evaluate the robustness of the ADAS feature. Selection of the baseline ratio for SPRT depends on the maturity level of the ADAS features and Operational Design Domain (ODD) / Object Event Detection Response (OEDR) coverage. Furthermore, SPRT utilizes the likelihood ratio approach to establish an acceptable, rejection and continuation regions. Number of hours/miles of accumulation and the number of mishaps/hazards are the two main factors for the robustness example shown in the paper. This paper demonstrates how to use these established regions to gain various levels of confidence and prove out the robustness of the ADAS features.</div></div>