Abstract: To evaluate the safety of takeover in Level 3 automated vehicles, the takeover scenario on a two-way six-lane highway is designed to conduct driving simulation experiments based on driving simulator. Drivers carry out the secondary tasks in vision all along during automated driving. The secondary tasks are watching arrowhead charts with three different-degrees of difficulty and the takeover scenario is that ago vehicle comes across a broken-down vehicle in front on the same lane, with a take-over request time of 7 s and an automated vehicle speed of 110 km/h. A total of 49 testees (30 males, 19 females) are recruited for experiment with an average age of 31.06 years (standard deviation: 7.10 years). After the vehicle sends auditory and visual request signals for takeover, testee should push the switch button on steering wheel to take back vehicle control. Bivariate logistic regression is used to build the evaluation model for takeover safety, with the group of minimum TTC taken as dependent variable: minimum TTC ≤ 1s is set as dangerous group and minimum TTC > 1s is set as safe group. The results show that in the condition of 7s takeover request time, the factors affecting takeover safety are takeover reaction time and secondary task, and the prediction accuracy of the takeover safety evaluation model built reaches 85.5%.

Key words: automated driving, driver behavior, takeover, modeling, secondary task