High-dimensional functional boundaries search for deviation-robust testing of autonomous driving system.

Area of Science:

• Autonomous Driving Systems
• Safety of the Intended Function (SOTIF)
• Systems Engineering

Background:

• Testing and evaluation are crucial for verifying the Safety of the Intended Function (SOTIF) in autonomous driving systems (ADS).
• Estimating system functional boundaries (SFB) requires designing effective safety-margin scenarios for test cases.
• Challenges include the curse of dimensionality, test coverage requirements, and surrogate model deviations in black-box testing.

Purpose of the Study:

• To propose an efficient framework for generating high-dimensional safety-margin scenarios for ADS.
• To develop a method for tracking the System Functional Boundary (SFB) of the system under test (SUT).
• To address the inaccuracies introduced by surrogate models in scenario generation and SFB estimation.

Main Methods:

• Utilizes a baseline surrogate model and a multi-population genetic algorithm (MPGA) to generate a diverse library of safety-margin test scenarios.
• Employs a System Functional Boundary Tracking (SFBT) module to compensate for deviations between the surrogate model and the actual SUT.
• Adaptively generalizes critical scenarios to estimate high-dimensional functional boundaries.

Main Results:

• The proposed framework efficiently generates high-dimensional safety-margin scenarios.
• The SFBT module effectively compensates for model deviations, leading to more accurate SFB estimation.
• Demonstrates a method for adaptive generalization of critical scenarios.

Conclusions:

• The developed framework offers an efficient approach to generating safety-margin scenarios for ADS.
• It improves the accuracy of System Functional Boundary (SFB) estimation by mitigating surrogate model deviations.
• This framework can aid in the testing and validation of the Operational Design Domain (ODD) for autonomous driving systems.