Development of an algorithm to identify stabilisation time for car-following after transitions of control from vehicle automation

The goal of this paper was to describe the development and validation of an algorithm able to detect the beginning of a car-following task engagement inside a time headway (THW) dataset. The motivation for this paper comes from the sensitivity of car-following models to noise inside the datasets, leading to unreliable results. Another aggravating factor for the noise inside such models is that nowadays, more studies are being developed considering the context of vehicle automation and transitions of control, where it is expected that drivers will have a certain delay until the time they recover motor coordination of the driving task and become able to follow a lead vehicle. The algorithm uses the concept of “stability” in car-following, which is defined as a constant but small fluctuation in vehicle’s THW as the criteria to identify the beginning of the task. In the end, a nested loop approach was applied, and the tool reached a performance of 89.2% reliability when tested against an experimental dataset, providing statistically significant improvement in the data quality by reducing unnecessary noise.