Predicting the duration of reduced driver performance during the automated driving takeover process

Abstract

This study carried out a simulator test to determine and predict the duration of reduced driver performance during the automated driving takeover process. Vehicle trajectory and driver behavior data were collected in critical and non-critical takeover scenarios. The earth mover’s distance was then adopted to identify the data with the optimal combination of indicators by comparing it to the reference data. The Gaussian mixture model was employed to classify the driving state as either stable or unstable, and the duration of reduced driver performance was derived for each participant based on these results. Subsequently, a generalized linear mixed model was developed to predict the duration of reduced driver performance and examine the impact of various factors on it. Results uncovered a recovery of the reduced driver performance state after drivers took over the automated vehicle. In the non-critical and critical takeover scenarios, the mean duration of reduced driver performance was 17.48 and 27.25 s, respectively. Additionally, the developed model demonstrated good overall prediction accuracy, with the duration of reduced driver performance showing a positive correlation with the lead vehicle’s speed, duration of automated driving, and takeover request lead time. Furthermore, timid drivers exhibited a longer recovery duration than aggressive drivers. These research findings offer valuable insights into understanding the recovery of reduced driver performance during the takeover process, serving as a theoretical foundation for designing safer automated driving systems.