Generating AV Test Scenarios Using AV Accident Data
Modeling of Autonomous Vehicle Crashes Using Open Measurable Scenario Description Language
Autonomous vehicle (AV) technology is expected to make a transformative impact on mobility. There are several companies already trying autonomous features of the vehicles in real traffic . Consequently, there have been crashes reported for these vehicles.
In this project, we integrate our validation methodology with Open Measurable Scenario Description Language (M-SDL) Developed by Foretellix Ltd. (https://www.foretellix.com/, https://www.calcalistech.com/ctech/articles/0,7340,L-3862892,00.html, https://www.prnewswire.com/il/news-releases/foretellix-closes-14m-series-a-funding-for-the-measurable-safety-and-broad-deployment-of-autonomous-vehicles-837456923.html) to use the information from these crashes for the validation of AVs, which is the most critical obstacle in their mass deployment. We also collaborate with the Tallinn University of Technology (TalTech) and generate tests for their autonomous campus shuttle. Our methodology together with the Coverage Driven Verification approach will assist in validation by recreating crashes and their cousin scenarios, measuring test completeness of the scenario & parameter space and iterating with the relevant tools to expose bugs and edge cases. It will also be instrumental in forming the requirements of the AV accident reports.
Some highlights:
We analyzed existing autonomous vehicle crashes (Tesla, Waymo, etc.). An older version of our accident database is here: Â https://s3.amazonaws.com/Ocoos_Sp_Content/3293/mapProject/index.html
We integrate our validation methodology with Open Measurable Scenario Description Language (M-SDL) developed by Foretellix Ltd. to use the information from these crashes for the validation of AVs.
We create variations of the crashes to create validation scenarios.
We use Foretify from Foretellix, which we believe will be a common tool in the autonomous vehicle world in the near future. We received training for it, currently 1 PhD and 2 undergrad students are using it.
The software connects with SUMO simulation tool for the visualization of modeled scenarios.
Additional to the reports for the industry partner, an important output of this project will be the produced publications.