Is it Good or Bad to Provide Driver Fatigue Warning During Take-Over in Highly Automated Driving?

Abstract

When and how the driver should be intervened to relieve the fatigue status in Society of Automotive Engineers Level 3 automated driving has raised numerous debates. In this paper, we identify the driver’s fatigue level according to the driver’s facial features before the system issues the take-over request (TOR), and then perform the fatigue warning (FW) intervention to investigate whether the driver would be better prepared for taking over the vehicle safely and efficiently. In a simulator-based study, we compared the driver’s driving performance under the condition of whether a FW intervention was provided before a possible TOR, with the aim to gain insights into the ability of fatigued drivers to regain manual control and situation awareness after automated driving. A FW + TOR condition was compared with a TOR-only condition using a within-subject design with 30 participants. Under the effect of FW, participants exhibited better take-over performance; for example, they gazed at the road and placed their hands on the steering wheel earlier before taking over the vehicle. Moreover, in the FW + TOR state, they also exhibited better take-over performance, with a shorter average braking reaction time and a higher average speed, which reflects that the driver’s speed is more stable during the take-over of the vehicle and that the take-over task can be completed with a smaller deceleration. Furthermore, it is concluded that the FW (5 s before TOR) + TOR mode has greater potential to increase the safety and acceptance of automated driving, and could reduce the driver’s safety risk in a fatigue state when driving automatically compared with the FW (10 s before TOR) + TOR mode. It is concluded that FW (5 s before TOR) + TOR mode has the potential to increase safety and acceptance of automated driving as compared with systems that provide only TORs. In addition, we need to design take-over scenarios and non-driving-related tasks with different levels of complexity to satisfy subsequent research. The fatigue detection method based on image-processing techniques also needs further improvement.