Optimization-Based Scenario Space Sampling for Performance Boundary Estimation of Driver Reference Models

Abstract

This paper presents a methodology to estimate the performance boundary of driver reference models across arbitrary driving scenario parameter spaces. A Gaussian Process Classifier (GPC) surrogate model is employed to approximate the boundary, trained using a Bayesian optimization approach. This enables accurate estimation with a limited number of simulation runs, making the method particularly suitable for computationally expensive scenarios. We demonstrate the effectiveness of the approach by accurately identifying the performance boundary of the reference driver model defined in UNECE Regulation No. 157 in a cut-in scenario. The resulting surrogate model enables performance prediction in untested regions, offering a complementary alternative to traditional techniques such as Monte Carlo simulations. When used in conjunction with methods to generate test scenarios for automated driving systems, this method can be used to benchmark the system against the reference model.

Authors 3

1. Jobst Beckmann Aachen Institute for Automotive Engineering

   RWTH Aachen University

   Affiliation as printed

   Institute for Automotive Engineering, RWTH Aachen University,Aachen,Germany

2. Jannes Dirksen Aachen Institute for Automotive Engineering

   RWTH Aachen University

   Affiliation as printed

   Institute for Automotive Engineering, RWTH Aachen University,Aachen,Germany

3. Lutz Eckstein Aachen Institute for Automotive Engineering

   RWTH Aachen University

   Affiliation as printed

   Institute for Automotive Engineering, RWTH Aachen University,Aachen,Germany