Automated vehicles require rigorous evaluation to ensure safety before being available. Scenario-based accelerated evaluation methods using importance sampling are promising approaches whose core problem is constructing a high-efficiency importance sampling distribution. Typical methods for constructing the importance sampling distribution are prone to fall into local optima with low evaluation efficiency due to the increased search space when solving for high-dimensional parameters. In this study, an accelerated evaluation framework integrating prior knowledge that constrains the search space is proposed. The original distributions of scenario variables and the importance of scenario variables to the occurrence of rare events are taken as prior knowledge. The prior knowledge is used to constrain the search space to candidate distributions with high evaluation efficiency. The candidate distribution with the highest evaluation efficiency is used as the importance sampling distribution to accelerate the safety evaluation of automated vehicles. The proposed framework is validated using the HighD naturalistic driving dataset in cut-in scenarios. Simulation results show that the proposed framework can improve the evaluation efficiency by three orders of magnitude compared to the crude Monte Carlo, demonstrating that prior knowledge can be successfully used to constrain the search space for constructing a high-efficiency importance sampling distribution.
Accelerated Evaluation Framework Integrating Prior Knowledge for Automated Vehicle Safety
Lect.Notes Mechanical Engineering
The IAVSD International Symposium on Dynamics of Vehicles on Roads and Tracks ; 2023 ; Ottawa, ON, Canada August 21, 2023 - August 25, 2023
2024-10-13
11 pages
Article/Chapter (Book)
Electronic Resource
English
Measuring automated vehicle safety : forging a framework
TIBKAT | 2018
|Integrating Naturalistic Insights in Objective Multi-Vehicle Safety Framework
ArXiv | 2024
|