We introduce a conceptually novel method for eye-movement signal analysis. The method is general in that it does not place severe restrictions on sampling frequency, measurement noise or subject behavior. Event identification is based on segmentation that simultaneously denoises the signal and determines event boundaries. The full gaze position time-series is segmented into an approximately optimal piecewise linear function in O(n) time. Gaze feature parameters for classification into fixations, saccades, smooth pursuits and post-saccadic oscillations are derived from human labeling in a data-driven manner. The range of oculomotor events identified and the powerful denoising performance make the method useable for both low-noise controlled laboratory settings and high-noise complex field experiments. This is desirable for harmonizing the gaze behavior (in the wild) and oculomotor event identification (in the laboratory) approaches to eye movement behavior. Denoising and classification performance are assessed using multiple datasets. Full open source implementation is included.

For nearly 20 years, looking at the tangent point on the road edge has been prominent in models of visual orientation in curve driving. It is the most common interpretation of the commonly observed pattern of car drivers looking through a bend, or at the apex of the curve. Indeed, in the visual science literature, visual orientation towards the inside of a bend has become known as “tangent point orientation”. Yet, it remains to be empirically established whether it is the tangent point the drivers are looking at, or whether some other reference point on the road surface, or several reference points, are being targeted in addition to, or instead of, the tangent point. Recently discovered optokinetic pursuit eye-movements during curve driving can provide complementary evidence over and above traditional gaze-position measures. This paper presents the first detailed quantitative analysis of pursuit eye movements elicited by curvilinear optic flow in real driving. The data implicates the far zone beyond the tangent point as an important gaze target area during steady-state cornering. This is in line with the future path steering models, but difficult to reconcile with any pure tangent point steering model. We conclude that the tangent point steering models do not provide a general explanation of eye movement and steering during a curve driving sequence and cannot be considered uncritically as the default interpretation when the gaze position distribution is observed to be situated in the region of the curve apex.

A major unresolved question in understanding visually guided locomotion in humans is whether actions are driven solely by the immediately available optical information (model-free online control mechanisms), or whether internal models have a role in anticipating the future path. We designed two experiments to investigate this issue, measuring spontaneous gaze behaviour while steering, and predictive gaze behaviour when future path information was withheld. In Experiment 1 participants (N = 15) steered along a winding path with rich optic flow: gaze patterns were consistent with tracking waypoints on the future path 1–3 s ahead. In Experiment 2, participants (N = 12) followed a path presented only in the form of visual waypoints located on an otherwise featureless ground plane. New waypoints appeared periodically every 0.75 s and predictably 2 s ahead, except in 25% of the cases the waypoint at the expected location was not displayed. In these cases, there were always other visible waypoints for the participant to fixate, yet participants continued to make saccades to the empty, but predictable, waypoint locations (in line with internal models of the future path guiding gaze fixations). This would not be expected based upon existing model-free online steering control models, and strongly points to a need for models of steering control to include mechanisms for predictive gaze control that support anticipatory path following behaviours.

Drift diffusion (or evidence accumulation) models have found widespread use in the modelling of simple decision tasks. Extensions of these models, in which the model’s instantaneous drift rate is not fixed but instead allowed to vary over time as a function of a stream of perceptual inputs, have allowed these models to account for more complex sensorimotor decision tasks. However, many real-world tasks seemingly rely on a myriad of even more complex underlying processes. One interesting example is the task of deciding whether to cross a road with an approaching vehicle. This action decision seemingly depends on sensory information both about own affordances (whether one can make it across before the vehicle) and action intention of others (whether the vehicle is yielding to oneself). Here, we compared three extensions of a standard drift diffusion model, with regards to their ability to capture timing of pedestrian crossing decisions in a virtual reality environment. We find that a single variable-drift diffusion model (S-VDDM) in which the varying drift rate is determined by visual quantities describing vehicle approach and deceleration, saturated at an upper and lower bound, can explain multimodal distributions of crossing times well across a broad range vehicle approach scenarios. More complex models, which attempt to partition the final crossing decision into constituent perceptual decisions, improve the fit to the human data but further work is needed before firm conclusions can be drawn from this finding.

In this paper we present and qualitatively analyze an expert driver's gaze behavior in natural driving on a real road, with no specific experimental task or instruction. Previous eye tracking research on naturalistic tasks has revealed recurring patterns of gaze behavior that are surprisingly regular and repeatable. Lappi (2016) identified in the literature seven “qualitative laws of gaze behavior in the wild”: recurring patterns that tend to go together, the more so the more naturalistic the setting, all of them expected in extended sequences of fully naturalistic behavior. However, no study to date has observed all in a single experiment. Here, we wanted to do just that: present observations supporting all the “laws” in a single behavioral sequence by a single subject. We discuss the laws in terms of unresolved issues in driver modeling and open challenges for experimental and theoretical development.

Moving in natural environments is guided by looking where you are going. When entering a bend, car drivers direct their gaze toward the inside of the curve, in the region of the curve apex. This behavior has been analyzed in terms of both "tangent point models," which posit that drivers are looking at the tangent point (TP), and "future path models," which posit that drivers are visually targeting a point on the desired trajectory or future path (FP). This issue remains unresolved, partly due to the challenge of representing the changing visual projection of the trajectory into the driver's field of view. This paper reports a study of naturalistic driving, in which the FP in the field of view is explicitly modeled, and the TP and reference points on the FP are simultaneously analyzed as potential gaze targets. We argue that traditional area-of-interest methods commonly interpreted as supporting the TP hypothesis are problematic when the interest is contrasting multiple gaze targets. This prompts a critical reassessment of the empirical case for the ubiquity of looking at the TP and the generality of the TP hypothesis as an account of where people look when they steer. As a basis for representing driver gaze behavior, the FP is an equally valid point of departure. There are no overwhelming theoretical or empirical reasons for favoring the TP models over the FP models.

When negotiating bends car drivers perform gaze polling: their gaze shifts between guiding fixations (GFs; gaze directed 1–2 s ahead) and look-ahead fixations (LAFs; longer time headway). How might this behavior change in autonomous vehicles where the need for constant active visual guidance is removed? In this driving simulator study, we analyzed this gaze behavior both when the driver was in charge of steering or when steering was delegated to automation, separately for bend approach (straight line) and the entry of the bend (turn), and at various speeds. The analysis of gaze distributions relative to bend sections and driving conditions indicate that visual anticipation (through LAFs) is most prominent before entering the bend. Passive driving increased the proportion of LAFs with a concomitant decrease of GFs, and increased the gaze polling frequency. Gaze polling frequency also increased at higher speeds, in particular during the bend approach when steering was not performed. LAFs encompassed a wide range of eccentricities. To account for this heterogeneity two sub-categories serving distinct information requirements are proposed: mid-eccentricity LAFs could be more useful for anticipatory planning of steering actions, and far-eccentricity LAFs for monitoring potential hazards. The results support the idea that gaze and steering coordination may be strongly impacted in autonomous vehicles.

Car following (CF) models used in traffic engineering are often criticized for not incorporating “human factors” well known to affect driving. Some recent work has addressed this by augmenting the CF models with the Task-Capability Interface (TCI) model, by dynamically changing driving parameters as function of driver capability. We examined assumptions of these models experimentally using a self-paced visual occlusion paradigm in a simulated car following task. The results show strong, approximately one-to-one, correspondence between occlusion duration and increase in time headway. The correspondence was found between subjects and within subjects, on aggregate and individual sample level. The long time scale aggregate results support TCI-CF models that assume a linear increase in time headway in response to increased distraction. The short time scale individual sample level results suggest that drivers also adapt their visual sampling in response to transient changes in time headway, a mechanism which isn’t incorporated in the current models.