AV/ADAS Safety-Critical Testing Scenario Generation from Vehicle Crash Data

2022-01-0104

03/29/2022

Features
Event
WCX SAE World Congress Experience
Authors Abstract
Content
This research leverages publicly available crash data to construct safety-critical scenarios focusing primarily on Level 3 Automated Driving Systems (ADS) safety assessment under highway driving conditions. NHTSA’s Crashworthiness Data System (CDS) has a rich dataset of representative crashes sampled from numerous Primary Sampling Units (PSUs) across the country. Each of these datasets includes the storyline, road geometry information, detailed description of actors involved in the crash, weather information, scene diagrams, crash images, and a myriad of other crash-specific details. The methodology adopted aims to generate critical scenarios from real-world driving to complement the existent regulatory tests for the validation of L3 ADS. For this work, a four-step approach was adopted to extract safety-critical scenarios from crash data. Firstly, a methodology was developed to filter crash cases relevant to the scope and resulting pdf files, and numerical crash data were downloaded from the CDS website. Then, the numerical data is utilized to characterize cases into crash categories. Thirdly, the numeric data and pdf data are used to combine multiple crash cases into fewer representative ‘logical scenarios’ which define the road network, actors, actor types, crash storyline encompassing a feasible parameter space. Finally, the generated logical scenarios are used to develop specific safety-critical scenarios. The key outcome of this research is a set of safety-critical scenarios based on real-world crash data, categorized according to specific ADS control objectives that can be utilized to validate ADS functionalities.
Meta TagsDetails
DOI
https://doi.org/10.4271/2022-01-0104
Pages
9
Citation
Kibalama, D., Tulpule, P., and Chen, B., "AV/ADAS Safety-Critical Testing Scenario Generation from Vehicle Crash Data," SAE Technical Paper 2022-01-0104, 2022, https://doi.org/10.4271/2022-01-0104.
Additional Details
Publisher
Published
Mar 29, 2022
Product Code
2022-01-0104
Content Type
Technical Paper
Language
English