Name
Title | ||
|---|---|---|
Accelerated Testing for Highly Automated Vehicles: A Combined Method Based on Importance Sampling and Normalizing Flows |
Abstract | ||
|---|---|---|
The risk rate of Highly Automated Vehicles (HAVs) represents their self-driving capability. One goal of safety testing is to estimate the risk rate of HAVs in naturalistic driving environment and compare them with human drivers. Due to the low exposure of risk events, such statistic estimation requires massive tests accumulation. To address this problem, we propose an enhanced accelerated testing method which combines the Importance Sampling and Normalizing Flows together. Importance Sampling can replace the naturalistic probability density function with a new one where rare events are more likely to occur. Considering that fitting Importance Sampling distribution with conventional parametric distributions is limited., the method of Normalizing Flows is introduced. It is capable to learn the distribution of rare events and to estimate Importance Sampling distribution by applying a sequence of invertible transformations. Normalizing Flows also maintains the correlated relationship between scenario variables. We apply it on naturalistic Car-following scenarios and estimate the collision rate of our System Under Test (SUT). It shows the combined method reduces the number of tests by 253 times compared to Monte Carlo when the relative half width reaches the required accuracy. The proposed method has great potential in accelerating the safety assessment of HAVs. |
Year | DOI | Venue |
|---|---|---|
2022 | 10.1109/ITSC55140.2022.9922218 | 2022 IEEE 25th International Conference on Intelligent Transportation Systems (ITSC) |
Keywords | DocType | ISBN |
normalizing flows,naturalistic car-following scenarios,collision rate,HAV,highly automated vehicles,risk rate,self-driving capability,safety testing,naturalistic driving environment,statistic estimation,massive tests accumulation,enhanced accelerated testing method,naturalistic probability density function,importance sampling distribution,conventional parametric distributions,combined method,invertible transformation,system under test,SUT,Monte Carlo method | Conference | 978-1-6654-6881-7 |
Citations | PageRank | References |
0 | 0.34 | 4 |
Authors | ||
3 | ||