This paper introduces a reference model of glance behavior for driving safety assessment. This model can improve the design of automated and assistive systems. Technological limitations have previously hindered the use of unobtrusive eye trackers to measure glance behavior in naturalistic conditions. This paper presents a comprehensive analysis of eye-tracking data collected in a naturalistic field operation test, using an eye tracker that proved to be robust in real-world driving scenarios. We describe a post-processing technique to enhance the quality of naturalistic eye-tracker data, propose a data-analysis procedure that captures the important features of glance behavior, and develop a model of glance behavior (based on distribution fitting), which was lacking in the literature. The model and its metrics capture key defining characteristics of, and differences between, on-and off-path glance distributions, during manual driving and driving with adaptive cruise control and lane keeping aid active. The results show that drivers' visual response is tightly coupled to the driving context (vehicle automation, car-following, and illumination).

Adaptive Cruise Control (ACC) has been shown to reduce the exposure to critical situations 12 by maintaining a safe speed and headway. It has also been shown that drivers adapt their visual behavior 13 in response to the driving task demand with ACC, anticipating an impending lead vehicle conflict by 14 directing their eyes to the forward path before a situation becomes critical. The purpose of this paper is 15 to identify the causes related to this anticipatory mechanism, by investigating drivers' visual behavior 16 while driving with ACC when a potential critical situation is encountered, identified as a forward 17 collision warning (FCW) onset (including false positive warnings). This paper discusses how sensory 18 cues capture attention to the forward path in anticipation of the FCW onset. The analysis used the 19 naturalistic database EuroFOT to examine visual behavior with respect to two manually-coded metrics, 20 glance location and glance eccentricity, and then related the findings to vehicle data (such as speed, 21 acceleration, and radar information). Three sensory cues (longitudinal deceleration, looming, and brake 22 lights) were found to be relevant for capturing driver attention and increase glances to the forward path 23 in anticipation of the threat; the deceleration cue seems to be dominant. The results also show that the 24 FCW acts as an effective attention-orienting mechanism when no threat anticipation is present. These 25 findings, relevant to the study of automation, provide additional information about drivers' response to 26 potential lead-vehicle conflicts when longitudinal control is automated. Moreover, these results suggest 27 that sensory cues are important for alerting drivers to an impending critical situation, allowing for a 28 prompt reaction. 29

We quantify the time-course of glance behavior and steering wheel control level in driver-initiated, non-critical disengagements of Tesla Autopilot (AP) in naturalistic driving. Although widely used, there are limited objective data on the impact of AP on driver behavior. We offer insights from 19 Tesla vehicle owners on driver behavior when using AP and transitioning to manual driving. Glance behavior and steering wheel control level were coded for 298 highway driving disengagements. The average proportion of off-road glances decreased from 36% when AP was engaged to 24% while driving manually after AP disengagement. Most of the offroad glances before the transition were downward and to the center stack (17%). Lastly, in 33% of the events drivers were not holding the steering wheel prior to AP disengagement. The study helps begin to enhance society's understanding, and provide a reference, of real-world AP use.

No abstract

Previous research indicates that drivers may forgo their supervisory role with partial-automation. We investigated if this behavior change is the result of the time automation was active. Naturalistic data was collected from 16 Tesla owners driving under free-flow highway conditions. We coded glance location and steering-wheel control level around Tesla Autopilot (AP) engagements, driver-initiated AP disengagements, and AP steady-state use in-between engagement and disengagement. Results indicated that immediately after AP engagement, glances downwards and to the center-stack increased above 18% and there was a 32% increase in the proportion of hands-free driving. The decrease in driver engagement in driving was not gradual over-time but occurred immediately after engaging AP. These behaviors were maintained throughout the drive with AP until drivers approached AP disengagement. In conclusion, drivers may not be using AP as recommended (intentionally or not), reinforcing the call for improved ways to ensure drivers’ supervisory role when using partial-automation.