Cut-Out Scenario Generation With Reasonability Foreseeable Parameter Range From Real Highway Dataset for Autonomous Vehicle Assessment

Abstract

This work was funded by the Ministry of Economy, Trade and Industry of Japan through the SAKURA project (https://www.sakura-prj.go.jp/).

A Quantitative Method to Determine What Collisions are Reasonably Foreseeable and Preventable

“…Similar as in (Nakamura et al, 2022;Muslim et al, 2023), both approaches presented in this work construct a PDF which is used to estimate the range of the parameters of the reasonably foreseeable scenarios.…”

“…Our proposed methods differ from the method used in (Nakamura et al, 2022;Muslim et al, 2023) because fewer assumptions regarding the PDF estimation are made. In (Nakamura et al, 2022;Muslim et al, 2023), it is assumed that the scenario parameters are independent, that the parameters are distributed according to the Beta distribution, and the lower and upper bounds of these Beta distributions are set to some assumed values. If we would use the method proposed in (Nakamura et al, 2022;Muslim et al, 2023), we would have reported a different outcome.…”

“…To answer this research question, we follow a similar approach as (Nakamura et al, 2022;Muslim et al, 2023), i.e., based on scenarios extracted from real-world data, we estimate the range of the scenario parameters that are reasonably foreseeable. To estimate this range, a Probability Density Function (PDF) of the scenario parameters is estimated.…”

“…The first approach employs the non-parametric PDF estimation technique called Kernel Density Estimation (KDE). Therefore, in contrast to the work of Nakamura et al (2022); Muslim et al (2023), no assumptions are needed regarding a parametric distribution and the independence of the scenario parameters. The second approach uses only part of the data that is above a certain threshold, such that the Generalized Pareto Distribution (GPD) can be assumed according to the Extreme Value Theory (EVT).…”

“…In (Kusano et al, 2022), the collision avoidance performance of the Waymo ADS is compared with that of a Non-Impaired Eyes ON the conflict (NIEON) reference behavior model. Nakamura et al (2022); Muslim et al (2023) provide a method to determine the range of the parameter values of "foreseeable" scenarios. This is done by fitting parametric distributions to the scenario parameter values and determining a range of the values that capture, e.g., 99 %, of the probability mass.…”