Evaluation of Driver Drowsiness While Using Automated Driving Systems on Driving Simulator, Test Course and Public Roads

摘要

This paper describes an investigation of evaluation indices for assessing driver conditions when using an automated driving system. We focused on a driver drowsiness in the automated mode. A driving simulator experiment was conducted to identify evaluation indices which were sensitive to the subjective evaluation of the driver's drowsiness. The following indices were calculated based on the driver's eye movement data recorded for 60 s before the Rtl (Request to Intervene): number of blinks, duration of blinking, PERCLOS (Percent of Eyelid Closure), pupil diameter, number of saccade, amplitude of saccade, and velocity of saccade. We also measured the driver's driving performance after a transition from the automated driving to the manual driving mode. The results of the driving simulator experiment suggested that PERCLOS was sensitive to the subjective assessment of the reduction of the driver's alert level. And this index was highly related to the time to initiate driver's steering operation after the Rtl presentation. We have developed a prototype of the driver monitoring system that detects drivers' eyelid movements. The findings obtained from a test course experiment and a public road experiment indicated the effectivity of the driver monitoring system for evaluating quantitatively the driver's drowsiness in the automated driving condition. The results of the public road experiment imply that the duration of blinking as well as PERCLOS might be necessary to estimate the delay of the steering response time after the transition to manual driving.