ADAS simulation project helps development of autonomous vehicle features

More and more frequently, new passenger cars are being equipped with advanced driver assistance systems (ADAS). Ensuring the safety of these systems is a big task that needs to be tackled with simulation. 

In a new project headed by SIMCenter researcher Punit Tulpule alongside graduate students Dennis Kibalama and Bo-Shian Chen, researchers used National Highway Safety Administration (NHTSA) accident records to create simulation scenarios to test level three autonomous ADAS features.  

Level three features include automatic lane keeping, obstacle avoidance, emergency braking, and abilities that allow the vehicle to perform maneuvers without the driver’s direction.  

Due to the wide variety of potential accident situations, the project uses a large pool of police report data collected by the NHTSA between 2004 and 2015 and focuses solely on highway situations. The goal is to create a set of test scenarios that encompasses all potential accident situations that could be experienced on the highway. 

In order to create the simulation scenarios, researchers converted accident data into information capable of being simulated. This was done by binning the NHTSA crashworthiness data into broad categories according to their crash types and storyline. The crash cases from each category were then analyzed to develop fewer logical scenarios. At this step the road network and agents were defined.  

To test the ADAS controllers, it was important to choose the right values of these parameters so that if the controller under test failed, a crash would occur, and it was feasible under the vehicle dynamics limits to avoid a crash. Finding these parameters in such a way that the resulting simulation scenario was sufficiently critical was done by using simple control assumptions and kinematic equations.  

In the end, 72 unique logical scenarios were identified, and for each logical scenario a few concrete scenarios were prepared. Each concrete scenario was simulated in Matlab® Scenario Builder Toolbox and OpenScenario standardized descriptor files were generated to have compatibility with other software tools. 

This simulation project will allow faster design cycles at lower costs and, most importantly, work toward improving the safety of future vehicles equipped with ADAS systems. 

“Autonomous vehicles are expected to be safer than human driving even under highly unlikely situations,” Tulpule said. “Testing these controllers in simulation could save human lives by letting the controller attempt various virtual situations.”