Methodology

Study participants engaged in a 40-minute interstate highway drive in a driving simulator with automated driving capability, while also being potentially distracted by a non-driving task. To evaluate driver performance, participants encountered three types of events during the drive, a freeze probe event to check driver awareness of surrounding vehicles, and events where the driver needed to take over from automated driving under both planned (approaching a work zone) and unplanned (sudden deactivation of automation) situations. Participants were divided into three groups to assess the AV human-machine interface (HMI):

• Baseline group – The system only provided alerts to drivers to take over driving in the event of a work zone ahead, or sudden automation deactivation.
• Attentional Maintenance (AM) group – The system also provided attentional reminders (e.g., “Look Forward”) when drivers looked away from the road for more than 30 seconds, and continued alerts until the driver looked back.
• State-contingent takeover (SCT) group – Drivers in this group were also given an earlier pre-alert notification shortly before the alerts, if the DMS sensed the driver was distracted.  

Participants’ situational awareness levels were used to assess whether the alerts improved driver performance when transitioning from automated control to driver control. Situational awareness was measured by percent road center gaze (visual attention to the roadway ahead) and freeze probe event accuracy (correctness in noting the locations of surrounding vehicles). Driver performance in takeover from automation was measured by the time to return their hands to the steering wheel and time to initiate a steering response, as well as the maximum lateral exceedance from the center of the lane.

Findings

• Providing attentional maintenance reminders increased drivers’ situational awareness and their ability to takeover during automation failures.
  • Percent road center gaze was higher for the AM group (35 percent) than the baseline group (15 percent).
  • Freeze probe accuracy was 80 percent for participants in the AM group and 73 percent in the baseline group.
  • Compared to the baseline group, the AM group were generally quicker to retake control of the steering wheel and to initiate steering.
  • Takeover performance as measured by maximum lateral exceedance from the center of the lane was statistically lower in the AM group as compared to baseline drivers. 
• No participants crashed in the work zone, suggesting that the takeover warnings (seven seconds in advance) provided an adequate amount of time for distracted drivers to re-engage with driving.
• Although state-contingent takeover requests improved some components of the takeover process, there was not a statistically significant difference in the takeover response of the participants in the SCT group and baseline group.