Authors:
Christoph Glasmacher
;
Hendrik Weber
;
Michael Schuldes
;
Nicolas Wagener
and
Lutz Eckstein
Affiliation:
Institute for Automotive Engineering, RWTH Aachen University, Aachen, Germany
Keyword(s):
Automated Driving, Intelligent Vehicles, Safety Assurance, Scenario Generation, Parameter Sampling, Causal Networks, Constraint Graphs.
Abstract:
Safety assurance of highly automated driving functions is a major challenge in today‘s research and requires the development of new validation methods. Scenario-based testing is a promising approach to handle the variety of possible situations efficiently. Due to the limited availability of real-world derived scenarios, they are increasingly generated synthetically. Whereas actual approaches to generate concrete parameters are mostly either knowledge- or data-driven, we propose a methodology to combine these approaches. We model the correlation of parameters in real-world data as multivariate probability functions by using copulas. In addition, we establish modular causal and constraint relations combining Bayesian networks and constraint graphs to add semantic knowledge about parameters and their interactions. Thereby, road user behavior and physical equations are represented. The application of our generation method on urban intersections shows the capability to sample high-dimensi
onal parameter spaces with limited input data. Hereby, it offers the opportunity to create realistic scenarios to extend the database for scenario-based assessment.
(More)