Communications of the ACM
Automated Technique to Discover Simulation Configurations for Behaviors Hard to Test
A research team of Japan's National Institute of Informatics developed a technique to search automatically for simulation configurations that test various behaviors of automated driving systems.
Credit: AdobeStock
Researchers at Japan's National Institute of Informatics (NII) have designed a technique to search automatically for simulation models that evaluate specific behaviors of automated driving systems.
The method iterates trials on simulations using evolutionary computation, in order to identify configurations that lead to specific features of driving behaviors like high acceleration, deceleration, and steering operation.
The technique avoids generating configurations that solely lead to dangerous situations, revealing features of driving behaviors not constrained to emergency situations.
In applying and assessing the method on a path-planning program offered by car manufacturer Mazda, the team learned the technique could produce specific behaviors that were rarely generated randomly.
NII's Fuyuki Ishikawa said the team established a holistic series of testing and debugging methods by adapting techniques for conventional program code, with the goal of finding solutions like "desirable tests" and "desirable fix actions."
From National Institute of Informatics (Japan)
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA
No entries found