Risk Quantification for Automated Driving Systems in Real-World Driving Scenarios

Abstract

The development of safety validation methods is essential for the safe deployment and operation of Automated Driving Systems (ADSs). One of the goals of safety validation is to prospectively evaluate the risk of an ADS dealing with real-world traffic. ISO 26262 and ISO/DIS 21448, the leading standards in automotive safety, provide an approach to estimate the risk where the former focuses on risks due to potential malfunctioning of components and the latter focuses on risks due to possible functional insufficiencies. The main shortcomings of the approach provided in ISO 26262 are that it depends on subjective judgments of safety experts and that only a qualitative risk estimation is performed. ISO/DIS 21448 addresses these shortcomings partially by providing statistical methods to guide the safety validation, but no complete method is provided to quantify the risk. The first objective of this article is to propose a method to estimate the risk of an ADS in a more quantitative and objective manner. A data-driven approach is used to rely less on subjective judgments of safety experts. The output of the method is the expected number of injuries in a potential collision. Thus, the method is quantitative, the result is easily interpretable, and the result can be compared with road crash statistics. The second objective is to provide a method that supports the risk assessment as stipulated by the ISO 26262 and ISO/DIS 21448 standards by decomposing the quantified risk into the 3 aspects of risk as mentioned in these standards: exposure, severity, and controllability. The proposed methods are illustrated by means of a case study in which the risk is quantified for a longitudinal controller in 3 different types of scenarios. The code of the case study is publicly available.

Deriving Environmental Risk Profiles for Autonomous Vehicles From Simulated Trips

“…The AV algorithm is then tested in a virtual environment where all possible combinations of this parameter values are simulated. For example, De Gelder et al ( 2019) [21] parameterize the environment, and calculate the exposure and severity of the AV for different scenario types (cut-inn during lane change etc.). This approach is the most relevant for the question that is being answered in this paper.…”

Infrastructure-Based Pedestrian Risk Tagging Methodology to Support AV Risk Assessment

“…Quantitative approaches include using trial AV operations to scan and gain operational data to learn and improve riskrelated issues [21] or using critical real-world scenarios [5,6,[22][23][24] to achieve likewise. In parallel, a growing new interest in using an independent real-time risk assessment [7][8][9][10][11] or monitoring device [25] is pursued to indicate real-time risk at the AV level.…”

“…Taking this into consideration, an additional standard ISO/PAS 21448 [3], also known as Safety of the Intended Functionality (SOTIF), was released to identify risks resulting from functional insufficiencies corresponding to both software and hardware performance limitations. This led to the exponential increase in scenariobased validation [4][5][6] extracted from real-world scenarios during development to form solutions for the identified risks. However, even with the combined scenario-based and traditional validation approach, gaps exist between the realworld and development considerations.…”

A Quantitative Method to Determine What Collisions are Reasonably Foreseeable and Preventable

“…In Section 2.2.1, we first estimate to which extent a skilled and attentive human driver can avoid collisions considering all scenarios within a scenario category. We will employ Monte Carlo simulations using importance sampling to estimate this probability (de Gelder et al, 2021). As an alternative approach, in Section 2.2.2, we estimate to which extent a skilled and attentive human driver can avoid a collision considering a specific scenario.…”