Takeover safety is a critical issue when using Level 2 advanced driver assistance systems. Understanding the context of takeover can facilitate the development of driver monitoring systems that can adapt to changing environments for more contextually appropriate assistance during takeover. The paper presents a hierarchical clustering analysis of hundreds of post-takeover vehicle kinematics in the MIT-AVT naturalistic driving study. Results show similar types of takeovers between Super Cruise and Autopilot: normal takeover, braking takeover, accelerating takeover, evasive-manoeuvre takeover, and right-swerve takeover (Autopilot only). Context analysis showed that braking takeover which occurred at a normal highway speed was often associated with upcoming highway exits and foreseeable low-speed situations, while accelerating, evasive-manoeuvre, and right-swerve takeovers were caused by strong brake (for Super Cruise) or large steering (for Autopilot) during slow car following. The findings indicate the potential for sensor-based approaches to assessing various contexts and facilitating a more holistic takeover reference model.

Advanced Driver Assistance Systems (ADAS) with proactive alerts have been used to increase driving safety. Such systems’ performance greatly depends on how accurately and quickly the risky situations and maneuvers are detected. Existing ADAS provide warnings based on the vehicle’s operational status, detection of environments, and the drivers’ overt actions (e.g., using turn signals or steering wheels), which may not give drivers as much as optimal time to react. In this paper, we proposed a spatio-temporal attention-based neural network to predict drivers’ lane-change intention by fusing the videos from both in-cabin and forward perspectives. The Convolutional Neural Network (CNN)-Recursive Neural Network (RNN) network architecture was leveraged to extract both the spatial and temporal information. On top of this network backbone structure, the feature maps from different time steps and perspectives were fused using multi-head self-attention at each resolution of the CNN. The proposed model was trained and evaluated using a processed subset of the MIT Advanced Vehicle Technology (MIT-AVT) dataset which contains synchronized CAN data, 11058-second videos from 3 different views, 548 lane-change events, and 274 non-lane-change events performed by 83 drivers. The results demonstrate that the model achieves 87% F1-score within the 1-second validation window and 70% F1-score within the 5-second validation window with real-time performance.

Non-intrusive, real-time analysis of the dynamics of the eye region allows us to monitor humans’ visual attention allocation and estimate their mental state during the performance of real-world tasks, which can potentially benefit a wide range of human-computer interaction (HCI) applications. While commercial eye-tracking devices have been frequently employed, the difficulty of customizing these devices places unnecessary constraints on the exploration of more efficient, end-to-end models of eye dynamics. In this work, we propose CLERA, a unified model for Cognitive Load and Eye Region Analysis, which achieves precise keypoint detection and spatiotemporal tracking in a joint-learning framework. Our method demonstrates significant efficiency and outperforms prior work on tasks including cognitive load estimation, eye landmark detection, and blink estimation. We also introduce a large-scale dataset of 30 k human faces with joint pupil, eye-openness, and landmark annotation, which aims at supporting future HCI research on human factors and eye-related analysis.

Developing vehicle automation that accommodates other road users and exhibits familiar behaviors may enhance traffic safety, efficiency, and fairness, leading to tolerance of the technology. However, the interdependence between vehicle automation and other road users makes them more challenging than typical control and path planning tasks. Through the lens of joint activity theory, we model driver and pedestrian behavior to explore how they balance and negotiate competing risk and velocity goals through movement. Joint activity theory informs an interpretation of these movements as signals, which can be associated with perceptual processes. We use simulation-based inference to estimate parameters of coupled driver and pedestrian perceptual models using naturalistic driving data. Perceptual models provide links between the processes guiding evaluation of risk and velocity maintenance, and how they govern driver acceleration and pedestrian walking. We found that the coupled simulations describe how drivers adjust their yielding behavior in the face of pedestrian risk, and how risk affects pedestrians’ decisions to cross. Dynamic risk and velocity parameters predicted safety, efficiency, and fairness outcomes, suggesting that the parameters and their dynamic perceptual models describe important components of the interactions. Traditional approaches employ static, summary predictors, which may fail to capture their continuous evolution and negotiation over time. Dynamic models of the interaction between drivers and pedestrians can inform vehicle automation by identifying deviations from communication norms, extracting interaction features, and evaluating communication and coordination.

Communication among road users smooths interactions, improves efficiency, and mitigates risk. Eye contact and waving may be the most salient of this communication, but more often road users use their movement or position as implicit signals. Vehicle automation may disrupt these signals by introducing unfamiliar or unclear interactions that may not align with other road user expectations. This creates a need to evaluate how effectively vehicle automation communicates and how this affects safety and efficiency. Vehicle automation with effective communication can enter the roadway ecology more naturally and facilitate acceptance across society. We modeled the outcomes of vehicle-pedestrian encounters where drivers yielded for pedestrians. Models of the initial conditions revealed that drivers and pedestrians jointly contribute to safety and efficiency; however, the initial conditions were generally poor predictors, suggesting that dynamic interaction may be an important determinant of those outcomes. An interdependence model of wait times revealed that driver and pedestrian influence on one another varied across traffic control devices. During the nonintersection encounters, pedestrian wait time depended on their own speed and distance when entering the encounter, indicating that they may linger away from the road and choose the encounter conditions. The stop sign encounters showed a division of influence, with pedestrian and driver initial conditions influencing their own wait time. The unprotected encounters showed negotiation, with pedestrian initial conditions influencing the driver wait time. We demonstrate the need and methods to understand interdependence between road users and the implications for vehicle automation communication.

Distracted drivers are dangerous drivers. Equipping advanced driver assistance systems (ADAS) with the ability to detect driver distraction can help prevent accidents and improve driver safety. In order to detect driver distraction, an ADAS must be able to monitor their visual attention. We propose a model that takes as input a patch of the driver's face along with a crop of the eye-region and classifies their glance into 6 coarse regions-of-interest (ROIs) in the vehicle. We demonstrate that an hourglass network, trained with an additional reconstruction loss, allows the model to learn stronger contextual feature representations than a traditional encoder-only classification module. To make the system robust to subject-specific variations in appearance and behavior, we design a personalized hourglass model tuned with an auxiliary input representing the driver's baseline glance behavior. Finally, we present a weakly supervised multi-domain training regimen that enables the hourglass to jointly learn representations from different domains (varying in camera type, angle), utilizing unlabeled samples and thereby reducing annotation cost.

Human visual perception forms different levels of abstractions expressing the essential semantic components in the scene at different scales. For real-world applications such as driving scene perception, abstractions of both coarse-level, such as the spatial presence of the lead vehicle, and the fine-level, such as the words on a traffic sign, serve as important signals for driver's decision making. However, the granularities of perception required for levels of abstractions are generally different. While current computer vision research makes significant progress in tasks of understanding the global scene (image classification), and dense scene (semantic segmentation), our work takes steps to explore the gap in between. In this paper, we propose Multi-class Probability Pyramid as a representation built on the top of pixel-level semantic scene labels. This representation forms region-level abstractions by controlling the granularity of local semantic information, and thus disentangles the variation of scene semantics at different resolutions. We further show how such representation can be effectively used for evaluation purposes, including interpretable evaluation of scene segmentation and unsupervised diagnosis of segmentation predictions.

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 off-road 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.

Solving the driving scene perception problem for driver-assistance systems and autonomous vehicles requires accurate and robust performance in both regularly-occurring driving scenarios (termed “common cases”) and rare outlier scenarios (termed “edge cases”). We propose an automated method for clustering common cases and detecting edge cases based on the visual characteristics of the external scene using deep learning. We apply this approach to develop a large-scale real-world video driving scene dataset of edge cases and common cases. This dataset consists of 1,156,592 10-second video clips, including 450 clusters of common cases, and 5,601 edge cases. We assign human-interpretable metadata labels (e.g., weather, lighting conditions) to the clusters through manual annotation. We further propose two automated methods for large-scale evaluation of scene segmentation models on naturalistic driving datasets that can capture potential system failures without human inspection. Video illustrations of select clusters will be made available to help with future research.

Cognitive load has been shown, over hundreds of validated studies, to be an important variable for understanding human performance. However, establishing practical, non-contact approaches for automated estimation of cognitive load under real-world conditions is far from a solved problem. Toward the goal of designing such a system, we propose two novel vision-based methods for cognitive load estimation, and evaluate them on a large-scale dataset collected under real-world driving conditions. Cognitive load is defined by which of 3 levels of a validated reference task the observed subject was performing. On this 3-class problem, our best proposed method of using 3D convolutional neural networks achieves 86.1% accuracy at predicting task-induced cognitive load in a sample of 92 subjects from video alone. This work uses the driving context as a training and evaluation dataset, but the trained network is not constrained to the driving environment as it requires no calibration and makes no assumptions about the subject's visual appearance, activity, head pose, scale, and perspective.

Vehicle technology naming has the potential to influence drivers' expectations (mental model) of the level of autonomous operation supported by semi-automated technologies that are rapidly becoming available in new vehicles. If divergence exists between expectations and actual design specifications, it may make it harder to develop trust or clear expectations of systems, thus mitigating potential benefits. Alternately, over-trust and misuse due to misunderstanding increase the potential for adverse events. An online survey investigated whether and how names of advanced driver assistance systems (ADAS) and automation features relate to expected automation levels. Systems with "Cruise" in their names were associated with lower levels of automation. "Assist" systems appeared to create confusion between whether the driver is assisting the system or vice versa. Survey findings indicate the importance of vehicle technology naming and its impact in influencing drivers' expectations of responsibility between the driver and system in who performs individual driving functions.

Voice interfaces offer promise in allowing drivers to keep their eyes on-road and hands on-wheel. In relieving visualmanual demand, there is the potential for voice-enabled interfaces to inadvertently shift the burden of load to cognitive resources. Measurement approaches are needed that can identify when and to what extent cognitive load is present during driving. A modified form of the AttenD algorithm was applied to assess the amount of cognitive load present in a set of auditory-vocal task interactions. These tasks were subset from a larger on-road study conducted in the Boston area of driver response during use of an in-vehicle voice system [22]. The modified algorithm differentiated among the set of auditory-vocal tasks examined -- and may be useful to HMI practitioners who are working to develop and evaluate HMIs to support drivers in managing their attention to the road, and in the development of real-time driver attention monitoring systems.

We consider a large dataset of real-world, on-road driving from a 100-car naturalistic study to explore the predictive power of driver glances and, specifically, to answer the following question: what can be predicted about the state of the driver and the state of the driving environment from a 6-second sequence of macro-glances? The context-based nature of such glances allows for application of supervised learning to the problem of vision-based gaze estimation, making it robust, accurate, and reliable in messy, real-world conditions. So, it's valuable to ask whether such macro-glances can be used to infer behavioral, environmental, and demographic variables? We analyze 27 binary classification problems based on these variables. The takeaway is that glance can be used as part of a multi-sensor real-time system to predict radio-tuning, fatigue state, failure to signal, talking, and several environment variables.

In-vehicle user interfaces increasingly rely on digital text to display information to the driver. Led by Apple's iOS, thin, lightweight typography has become increasingly popular in cutting-edge HMI designs. The legibility trade-offs of lightweight typography are sparsely studied, particularly in the glance-like reading scenarios necessitated by driving. Previous research has shown that even relatively subtle differences in the design of the on-screen typeface can influence to-device glance time in a measurable and meaningful way. Here we investigate the relative legibility of four different weights (line thicknesses) of type under two different rendering systems (suboptimal rendering and optimal rendering). Results indicate that under suboptimal rendering, the lightest weight typeface renders poorly and is associated with markedly degraded legibility. Under optimal rendering, lighter weight typefaces show enhanced legibility compared to heavier typefaces. The reasons for this pattern of results, and its implications for design considerations in modern HMIs, are discussed.

This study explores the effects of minor changes in automation level on drivers' engagement in secondary activities. Three levels of automation were tested: manual, semi-autonomous, and fully-autonomous. Potential distractor items were present and participants were instructed they could use them if they felt it was safe. Hand positions and engagement in secondary activities were manually coded. Participants were significantly less likely to engage in a secondary activity in semi-autonomous than fully-autonomous mode. Likewise, they were significantly less likely to use two hands to interact with a secondary activity in semi-autonomous mode than fully-autonomous mode. Gaze classification for each of the driver roles revealed that increasing levels of automation resulted in an increasing percentage of off-road glance durations. These observations suggest that in the event of automation failures, a driver in semi-autonomous driving may be in a somewhat better position to retake control and avoid collisions than during fully autonomous driving.

We introduce a recurrent neural network architecture for automated road surface wetness detection from audio of tire-surface interaction. The robustness of our approach is evaluated on 785,826 bins of audio that span an extensive range of vehicle speeds, noises from the environment, road surface types, and pavement conditions including international roughness index (IRI) values from 25 in/mi to 1400 in/mi. The training and evaluation of the model are performed on different roads to minimize the impact of environmental and other external factors on the accuracy of the classification. We achieve an unweighted average recall (UAR) of 93.2% across all vehicle speeds including 0 mph. The classifier still works at 0 mph because the discriminating signal is present in the sound of other vehicles driving by.

Accurate, robust, inexpensive gaze tracking in the car can help keep a driver safe by facilitating the more effective study of how to improve (i) vehicle interfaces and (ii) the design of future advanced driver assistance systems. In this study, the authors estimate head pose and eye pose from monocular video using methods developed extensively in prior work and ask two new interesting questions. First, how much better can they classify driver gaze using head and eye pose versus just using head pose? Second, are there individual‐specific gaze strategies that strongly correlate with how much gaze classification improves with the addition of eye pose information? The authors answer these questions by evaluating data drawn from an on‐road study of 40 drivers. The main insight of the study is conveyed through the analogy of an ‘owl’ and ‘lizard’ which describes the degree to which the eyes and the head move when shifting gaze. When the head moves a lot (‘owl’), not much classification improvement is attained by estimating eye pose on top of head pose. On the other hand, when the head stays still and only the eyes move (‘lizard’), classification accuracy increases significantly from adding in eye pose. The authors characterise how that accuracy varies between people, gaze strategies, and gaze regions.

A Samsung Galaxy S4 and Apple iPhone 5s were compared in a driving simulator where participants performed visual-manual and auditory-vocal address entry tasks. Auditory-vocal tasks were associated with shorter task times, fewer off-road glances, lower workload ratings, and reduced impact on vehicle performance. Primarily nominal differences were found between devices. Older participants had more difficulty performing tasks across both modalities, and difficulties were amplified for visual-manual tasks. A pattern of performance advantages were found when participants used the same operating system as the one they personally owned, including faster task completion, fewer off-road glances, and lower reported workload coupled with lower impact on lane variability. One's familiarity with a smartphone may, to some extent, impact the apparent level of demand associated with its use while driving. These findings highlight a need for a more comprehensive understanding of how technology training can aid in minimizing interface demands.

The work proposes conceptual ways of considering drivers' visual behavior over time as a flow of attention across space. These efforts extend upon previous works (research and regulatory) that focus in greater detail on quantifying attention independent of time. The primary contribution of this work lies in the presented methodology for analysis of glance allocation features. An application of this method is explored as an illustration of its analytic potential. Driver glance allocation data were drawn from a heterogeneous set of drivers. Double coded and mediated glance allocations were used to detail differences between drivers' visual behavior across 3 different task types (baseline driving, visual-manual interaction, auditory-vocal interaction). Glance transition counts were used to estimate glance transition probabilities and significance values. Glance duration features were explored in parallel and used to visualize temporal allocation distribution and significance. Results show that visualization techniques based on these features are well suited for qualitative analysis of visual attention.

Communicational signals (e.g. lights and horns) are imperative for on-road interaction between drivers. The aim of the present study was to explore how these signals affect drivers' subjective appraisal and visual attention, and how drivers decode the signals from other vehicles within a variety of interactive contexts. Twenty-five male participants (20 valid samples, ranging from 21 to 29 years of age) were recruited to watch film clips of pre-designed interactive scenarios involving common vehicle signals in a full-view simulator (i.e. including road view and mirror views). Participants' attitudes towards the interacting vehicle's behaviours, emotional responses, fixation metrics, and decoded meanings were recorded and analysed. The majority of tested signals, with the exception of the horn used in the behind vehicles, significantly improved drivers' attitudes and pleasure. All signals significantly increased emotional arousal, as well as the total fixation time and mean fixation duration on the interacting vehicle. When the interacting vehicle was visible in mirrors, the signal usage significantly increased the fixation frequency towards it. Meanwhile, a significant decrease in total fixation time and mean fixation duration on the road was reported. The results also demonstrated that the decoded signal contained several meanings simultaneously depending on both the signal type and its interactive context. This study quantified the communication process via vehicular signals under typical situations involving other vehicles, and also suggested new ideas on how to establish more advanced communication between drivers.

Associations between user characteristics and system features to technology adoption have been discussed in various domains. However, less is known about how different factors potentially affect the adoption of in-vehicle smart technologies. This study builds and tests a research model that describes the relationships of individual characteristics, preconceptions, and task performance and perceptions measured during a system experience to attitudes and expectations toward in-vehicle technologies. Based on empirical data from three research cases—voice-control interface, active parallel parking assist, and cross traffic alert—this study finds perceptions of a hands-on system experience to have strong associations with postexperience attitudes and expectations. Individual characteristics including age and health, general preconceptions, and task performance were found to have weaker relationships. Based on the findings, this article discusses implications for research in the emerging domain of smart technologies in automobiles, as well as for practice in design and delivery of in-vehicle technologies.

In-vehicle user interfaces increasingly rely on screens filled with digital text to display information to the driver. As these interfaces have the potential to increase the demands placed upon the driver, it is important to design them in a way that minimizes attention time to the device and thus keeps the driver focused on the road. Previous research has shown that even relatively subtle differences in the design of the on-screen typeface can influence to-device glance time in a measurable and meaningful way. Here we outline a methodology for rapidly and flexibly investigating the legibility of typefaces in glance-like contexts, and apply this method to a comparison of 5 Simplified Chinese typefaces. We find that the legibility of the typefaces, measured as the minimum presentation time needed to read character strings and respond to a yes/no lexical decision task, is sensitive to differences in the typeface's design characteristics. The most legible typeface under study could be read 33.1% faster than the least legible typeface in this glance-induced context. Benefits and limitations of the methodology are discussed.

A driving simulation study was performed to compare visual-manual (touch screen based) destination entry using a Samsung Galaxy S4 smartphone with the standard voice command based interface and a voice based "Hands-Free mode" that appears to be intended for use while driving (i.e. has a steering wheel icon adjacent to the mode selection menu and the voice interface menu screen is visually austere when compared with the standard voice mode). The performance of 24 drivers on an alphanumeric street address entry task was assessed with respect to subjective workload, task duration, standard deviation of lateral lane position, response to a detection response task (DRT), and heart rate. With the exception of heart rate, all evaluation measures indicate that the voice interfaces provide significant advantages over the touch interface. Furthermore, subjective workload ratings and task duration measures imply that the "Hands-Free" voice based mode may have some costs relative to the standard voice command based interface. Lastly, all destination entry methods were associated with an increased DRT reaction time and higher miss-rates compared to a baseline driving condition.

Multi-function in-vehicle interfaces are an increasingly common feature in automobiles. Over the past several years, these interfaces have taken on an ever-greater number of functions and the ways in which drivers interact with information have become more complex. Parallel with these technical developments, interest in ensuring that these systems minimize demand placed upon the driver has also increased. Voice command capability has become a popular and desirable feature, as interacting with a vehicle interface through auditory/vocal interactions is often hypothesized to allow the driver to keep their eyes on the road and hands on the wheel. However, research has shown that production level voice command systems may still impart considerable visual demands on the driver [18]. These demands might be due in part to screen displays associated with extensive confirmatory dialogue and the driver's desire for visual confirmation that commands were accurately recognized. This study extends this work by comparing the default mode of a production voice system with an "Expert" mode which streamlines tasks by removing several confirmatory steps. We found that, although the use of the Expert mode significantly reduces overall task completion time, it has no appreciable effect on the amount of visual engagement; drivers still glance off the road for durations that are consistent with the Default mode. Implications for interface design and driver safety are discussed.

Understanding the driver's cognitive load is important for evaluating in-vehicle user interfaces. This paper describes experiments to assess machine learning classification algorithms on their ability to automatically identify elevated cognitive workload levels in drivers, leading towards the development of robust tools for automobile user interface evaluation. We look at using both driver performance as well as physiological data. These measures can be collected in real-time and do not interfere with the primary task of driving the vehicle. We report classification accuracies of up to 90% for detecting elevated levels of cognitive load, and show that the inclusion of physiological data leads to higher classification accuracy than vehicle sensor data evaluated alone. Finally, we show results suggesting that models can be built to classify cognitive load across individuals, instead of building individual models for each per-son. By collecting data from drivers in two large field studies on the highway (20 drivers and 99 drivers), this work extends prior work and demonstrates feasibility and potential of such measures for HCI research in vehicles.

Workload is a dynamic concept that can have different meanings depending on the investigative perspective and the question being asked. Awareness of these methodological distinctions is essential when interpreting research and human machine interface evaluations. Understanding workload within the contexts of objective demand vs. effective workload, physical vs. mental workload, task pacing, automation, technology education, measurement context and distraction are among the themes raised in this brief review.

This paper reports on the initial results of an exploratory study of the impact of typeface design on glance behavior away from the roadway when a driver interacts with a multi-line menu display designed to model a text rich automotive human machine interface (HMI). Data from 42 participants ranging from 36 to 75 years of age was collected in a driving simulation experiment in which participants were asked to respond to a series of address, restaurant identification, and content search menus that were implemented using two different typeface designs. Among men, a "square grotesque" typeface resulted in a 12.2% increase in visual demand as compared to the "humanist" typeface. Total glance time and number of glances required to complete a response showed consistent results. This research suggests that optimizing typeface characteristics may be viewed as a "low cost" method of providing a significant reduction in interface demand and associated distractions. Future work will need to assess if other font characteristics can be tuned to provide further reductions in demand.

This study compared the performance metrics of drivers who carried out visual-manipulative and auditory in-vehicle tasks while driving. Although these two different types of secondary tasks resulted in similar levels of self-reported workload, performing the visual tasks had a much greater impact on measurements of lateral control and resulted in greater compensatory behavior. While performing the auditory tasks, the overlap in drivers' processing resources was less than that of the visual task. However, competition over cognitive (or central) resources was revealed through the drivers' eye movements. A reduction in the allocation of visual attention, as observed through the concentration of gaze while performing the auditory tasks, suggests that tasks involving an increase in cognitive workload can also impact driving performance. The impact on performance can cause safety concerns as drivers' compensatory adjustments have been observed to result in slower reaction times. These findings are consistent with the Multiple Resources Theory [1] and provide some quantitative support for the theory.

In this paper, we describe changes in heart rate and skin conductance that result from an artificial manipulation of driver cognitive workload during an on-road driving study. Cognitive workload was increased systematically through three levels of an auditory delayed digit recall (n-back) task. Results show that changes in heart rate and skin conductance with increasing levels of workload are similar to those observed in an earlier simulation study. Heart rate increased in a step-wise fashion through the first two increases in load and then showed a less marked increase at the highest task level. Skin conductance increased most dramatically during the first level of the cognitive task and then appeared to more rapidly approach a ceiling (leveling) than heart rate. Findings further demonstrate the applicability of physiological indices for detecting changes in driver workload.

The goal of this study was to assess heart rate and driving performance while middle age and younger adults engaged in a naturalistic hands free phone task that was structured to place objectively equivalent cognitive demands on all participants. Although heart rate measures have been used in evaluating driver workload, prior studies had not compared responses in middle age and younger adults. Younger and middle age subjects performed equally well on the cellular telephone task. Middle age subjects drove more slowly overall and, as a group, did not demonstrate heart rate acceleration in response to the phone conversation that was seen in younger drivers. Both age groups showed a drop in speed control during the task. Late middle age adults appear as capable as young adults of managing the additional workload of a low to moderately demanding cognitive task while driving.

It is now evident from anecdotal evidence and preliminary research that distractions can hinder the task of operating a vehicle, and consequently reduce driver safety. However with increasing wireless connectivity and the portability of office devices, the vehicle of the future is visualized as an extension of the static work place - i.e. an office-on-the-move, with a phone, a fax machine and a computer all within the reach of the vehicle operator. For this research a Head mounted Eye-tracking Device (HED), is used for tracking the eye movements of a driver navigating a test route in an automobible while completing various driving tasks. Issues arising from data collection of eye movements during the completion of various driving tasks as well as the analysis of this data are discussed. Methods for collecting video and scan-path data, as well as difficulties and limitations are also reported.