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

Abstract

This study aims to generate test cases for scenario-based assessment of automated driving systems (ADS) when encountering a cut-out maneuver where the lead vehicle having changed lanes, revealing a new lead vehicle that, in some cases, is slower than the original lead (the cutting-out) vehicle. We extracted the cut-out scenarios from an established real-world traffic dataset recorded by instrumented vehicles on Japanese highways and then defined using vehicle kinematic parameters (velocities and distances). The extracted scenarios were analyzed based on the direct correlation between every two consecutive vehicles: a rear part that describes the correlation between the following vehicle and the cutting-out vehicles; and a frontal part that describes the correlation between the cutting-out vehicle and the preceding vehicle. Parameter ranges were quantified with a regression model and determined based on the risk acceptance threshold applied in the field of Japanese high-speed trains and annual exposure by professional highway drivers to produce a scenario space with a reasonably foreseeable range in which ADS may not produce crashes lest it performs worse than human drivers. A multi-dimensional distribution analytical approach was used to derive a correlation between the following vehicle and the preceding vehicle considering the initial longitudinal velocities. Results suggest that there should be no risk of a collision between the following vehicle and the preceding vehicle when the time headway between the following vehicle and the cutting-out vehicle is equal to or more than 2 s. These findings can help to understand normative driver behavior during cut-out scenarios and to generate accident-free scenario space for which ADS must perform flawlessly.