Objective:The aim of this study was to understand how to secure driver supervision engagement and conflict intervention performance while using highly reliable (but not perfect) automation.Background:Securing driver engagement—by mitigating irony of automation (i.e., the better the automation, the less attention drivers will pay to traffic and the system, and the less capable they will be to resume control) and by communicating system limitations to avoid mental model misconceptions—is a major challenge in the human factors literature.Method:One hundred six drivers participated in three test-track experiments in which we studied driver intervention response to conflicts after driving highly reliable but supervised automation. After 30 min, a conflict occurred wherein the lead vehicle cut out of lane to reveal a conflict object in the form of either a stationary car or a garbage bag.Results:Supervision reminders effectively maintained drivers’ eyes on path and hands on wheel. However, neither these reminders nor explicit instructions on system limitations and supervision responsibilities prevented 28% (21/76) of drivers from crashing with their eyes on the conflict object (car or bag).Conclusion:The results uncover the important role of expectation mismatches, showing that a key component of driver engagement is cognitive (understanding the need for action), rather than purely visual (looking at the threat), or having hands on wheel.Application:Automation needs to be designed either so that it does not rely on the driver or so that the driver unmistakably understands that it is an assistance system that needs an active driver to lead and share control.

Advanced driver assistance systems should facilitate and possibly boost drivers' self-regulating behavior. For instance, they might recognize when appropriate adaptive behavior is missing and advise or alert accordingly. The results from this study could also inspire training programs for novice drivers, or locally classify roads in terms of the risk associated with secondary task engagement while driving.

a b s t r a c tNaturalistic driving studies show that drivers engaged in complex visual-manual tasks face an increased risk of a crash or near-crash. Tasks that require many glances and a high proportion of long glances away from the road are of special concern for safety. Driving context (e.g. turning maneuvers, presence of lead or oncoming vehicles, vehicle speed) may also influence drivers' glance behavior during normal driving, since the drivers may have to estimate curvature and anticipate potential threats (e.g., lead vehicle braking). However, the effect of driving context on glance behavior during visual-manual tasks has not yet been thoroughly investigated in naturalistic driving. The extent to which drivers adapt their glance behavior to changes in the road environment during secondary tasks is likely to influence their ability to compensate for and respond to changes in the road environment. The present study investigated for the first time the effect of both driving context and visual-manual phone tasks (i.e., dialing, texting, reading) on drivers' glance behavior in naturalistic driving.This study shows that drivers indeed spend more time looking at the road and have a lower proportion of long off-road glances in complex driving contexts such as when turning and when lead or oncoming vehicles are present, both in normal driving and while engaged in a visual-manual phone task. In particular, these findings are more pronounced during turning maneuvers and in the presence of oncoming vehicles than in the presence of lead vehicles. Interestingly, driving speed influenced off-road glance durations during the phone tasks, but not during normal driving.The results from this study highlight the need to take driving context into account when evaluating the influence of different secondary tasks, in-vehicle user interfaces and glance metrics on driving safety, including the risk of crash involvement. The finding that glance behavior is context-dependent in a naturalistic setting has further implications for distraction detection algorithms, driver support systems, and driver training. Finally, driving contexts should be matched when comparing glance behavior, while driving with and without secondary tasks.