Abstract:

When the driver is not required to supervise the vehicle at all times during the L3 autonomous system operation， he or she is often removed from the driving task and may be engaged in a variety of non-driving related tasks. When the autonomous driving system encounters unexpected situation and sends a takeover request， whether the driver can safely and timely conduct human-machine interaction and take over the vehicle is an important issue for L3 autonomous driving. In this paper， five different road scenarios are designed for L3 autonomous driving takeover experiments based on the difference of road alignment， the entropy of driver's gaze in different road scenarios is analyzed， a Markov chain gaze model of driver's takeover behavior is constructed， and the visual transfer characteristics of the driver during human-computer interaction in the intelligent cockpit are explored. The results show that the driver shows obvious staring behavior in the process of take over， with a focus on the road ahead， sub task areas and human-machine interaction areas. With the decrease of road curvature radius， the driver's staring behavior changes obviously， and the driver's attention to the front and right side of the road increases， while the attention to the sub-mission area decreases. The results can provide a scientific basis for the optimization of the human-machine interface for autonomous driving， thereby improving driver takeover performance and driving safety.

Key words: autonomous driving, human-computer interaction, human-machine takeover, gaze characteristics, Markov chain