Resolving uncertainty on the fly: modeling adaptive driving behavior as active inference.

Understanding adaptive human driving behavior, in particular how drivers manage uncertainty, is of key importance for developing simulated human driver models that can be used in the evaluation and development of autonomous vehicles. However, existing traffic psychology models of adaptive driving behavior either lack computational rigor or only address specific scenarios and/or behavioral phenomena. While models developed in the fields of machine learning and robotics can effectively learn adaptive driving behavior from data, due to their black box nature, they offer little or no explanation of the mechanisms underlying the adaptive behavior. Thus, generalizable, interpretable, computational models of adaptive human driving behavior are still rare. This paper proposes such a model based on active inference, a behavioral modeling framework originating in computational neuroscience. The model offers a principled solution to how humans trade progress against caution through policy selection based on the single mandate to minimize expected free energy. This casts goal-seeking and information-seeking (uncertainty-resolving) behavior under a single objective function, allowing the model to seamlessly resolve uncertainty as a means to obtain its goals. We apply the model in two apparently disparate driving scenarios that require managing uncertainty, (1) driving past an occluding object and (2) visual time-sharing between driving and a secondary task, and show how human-like adaptive driving behavior emerges from the single principle of expected free energy minimization.

Driving among individuals with chronic conditions: A systematic review of applied research using kinematic driving sensors.

BACKGROUND: Kinematic driving data studies are a novel methodology relevant to health care, but prior studies have considerable variance in their methods, populations, and findings suggesting a need for critical analysis and appraisal for feasibility and methodological guidelines. METHODS: We assessed kinematic driving studies of adults with chronic conditions for study feasibility, characteristics, and key findings, to generate recommendations for future study designs, and to identify promising directions for applications of kinematic driving data. PRISMA was used to guide the review and searches included PubMed, CINAHL, and Compendex. Of 379 abstract/titles screened, 49 full-text articles were reviewed, and 29 articles met inclusion criteria of analyzing trip-level kinematic driving data from adult drivers with chronic conditions. RESULTS: The predominant chronic conditions studied were Alzheimer's disease and related Dementias, obstructive sleep apnea, and diabetes mellitus. Study objectives included feasibility testing of kinematic driving data collection in the context of chronic conditions, comparisons of simulation with real-world kinematic driving behavior, assessments of driving behavior effects associated with chronic conditions, and prognostication or disease classification drawn from kinematic driving data. Across the studies, there was no consensus on devices, measures, or sampling parameters; however, studies showed evidence that driving behavior could reliably differentiate between adults with chronic conditions and healthy controls. CONCLUSIONS: Vehicle sensors can provide driver-specific measures relevant to clinical assessment and interventions. Using kinematic driving data to assess and address driving measures of individuals with multiple chronic conditions is positioned to amplify a functional outcome measure that matters to patients.

Automated vehicles for older adults with cognitive impairment: a survey study.

As the population is ageing, the number of older adults with cognitive impairment (CI) is increasing. Automated vehicles (AVs) can improve independence and enhance the mobility of these individuals. This study aimed to: (1) understand the perception of older adults (with and without CI) and stakeholders providing services and supports regarding care and transportation about AVs, and (2) suggest potential solutions to improve the perception of AVs for older adults with mild or moderate CI. A survey was conducted with 435 older adults with and without CI and 188 stakeholders (e.g. caregivers). The results were analysed using partial least square - structural equation modelling and multiple correspondence analysis. The findings suggested relationships between older adults' level of cognitive impairment, mobility, knowledge of AVs, and perception of AVs. The results provided recommendations to improve older adults' perception of AVs including education and adaptive driving simulation-based training.Practitioner summary: This study investigated the perception of older adults and other stakeholders regarding AVs. The findings suggested relationships between older adults' level of cognitive impairment, mobility, knowledge of AVs, and perception of AVs. The results provided guidelines to improve older adults' perception of AVs.

Driving decisions after critical illness: Qualitative analysis of patient-provider reviews during ICU recovery clinic assessments.

BACKGROUND: Driving a vehicle is a functional task requiring a threshold of physical, behavioral and cognitive skills. OBJECTIVE: To report patient-provider evaluations of driving status and driving safety assessments after critical illness. DESIGN: Qualitative secondary analysis of driving-related dialog drawn from a two-arm pilot study evaluating telemedicine delivery of Intensive Care Unit Recovery Clinic assessments. Multidisciplinary providers assessed physical, psychological, and cognitive recovery during one-hour telemedicine ICU-RC assessments. Qualitative secondary analysis of patient-provider dialog specific to driving practices after critical illness. SETTING AND PATIENTS: Multidisciplinary Intensive Care Unit Recovery clinic assessment dialog between 17 patients and their providers during 3-week and/or 12-week follow-up assessments at a tertiary academic medical center in the Southeastern United States. MAIN MEASURES AND KEY RESULTS: Thematic content analysis was performed to describe and classify driving safety discussion, driving status and driving practices after critical illness. Driving-related discussions occurred with 15 of 17 participants and were clinician-initiated. When assessed, driving status varied with participants reporting independent decisions to resume driving, delay driving and cease driving after critical illness. Patient-reported driving practices after critical illness included modifications to limit driving to medical appointments, self-assessments of trip durations, and inclusion of care partners as a safety measure for new onset fatigue while driving. CONCLUSION: We found that patients are largely self-navigating this stage of recovery, making subjective decisions on driving resumption and overall driving status. These results highlight that driving status changes are an often underrecognized yet salient social cost of critical illness. TRIAL REGISTRATION: Clinicaltrials.gov: NCT03926533.

Return to Driving After Critical Illness.

This cross-sectional study examines the postintensive care syndrome in patients who had vs patients who had not resumed driving 1 month after hospitalization for a critical illness.

Analysis of Silver Alert Reporting System Activations for Missing Adults With Dementia in Texas, 2017 to 2022.

This cross-sectional study analyzes data from Silver Alert activations in Texas from 2017 to 2022 to identify temporal, geographic, and wandering characteristics of missing adults with dementia.

A Bayesian Regression Analysis of the Effects of Alert Presence and Scenario Criticality on Automated Vehicle Takeover Performance.

OBJECTIVE: This study investigates the impact of silent and alerted failures on driver performance across two levels of scenario criticality during automated vehicle transitions of control. BACKGROUND: Recent analyses of automated vehicle crashes show that many crashes occur after a transition of control or a silent automation failure. A substantial amount of research has been dedicated to investigating the impact of various factors on drivers' responses, but silent failures and their interactions with scenario criticality are understudied. METHOD: A driving simulator study was conducted comparing scenario criticality, alert presence, and two driving scenarios. Bayesian regression models and Fisher's exact tests were used to investigate the impact of alert and scenario criticality on takeover performance. RESULTS: The results show that silent failures increase takeover times and the intensity of posttakeover maximum accelerations and decrease the posttakeover minimum time-to-collision. While the predicted average impact of silent failures on takeover time was practically low, the effects on minimum time-to-collision and maximum accelerations were safety-significant. The analysis of posttakeover control interaction effects shows that the effect of alert presence differs by the scenario criticality. CONCLUSION: Although the impact of the absence of an alert on takeover performance was less than that of scenario criticality, silent failures seem to play a substantial role-by leading to an unsafe maneuver-in critical automated vehicle takeovers. APPLICATION: Understanding the implications of silent failure on driver's takeover performance can benefit the assessment of automated vehicles' safety and provide guidance for fail-safe system designs.

Predicting Procedure Step Performance From Operator and Text Features: A Critical First Step Toward Machine Learning-Driven Procedure Design.

OBJECTIVE: The goal of this study is to assess machine learning for predicting procedure performance from operator and procedure characteristics. BACKGROUND: Procedures are vital for the performance and safety of high-risk industries. Current procedure design guidelines are insufficient because they rely on subjective assessments and qualitative analyses that struggle to integrate and quantify the diversity of factors that influence procedure performance. METHOD: We used data from a 25-participant study with four procedures, conducted on a high-fidelity oil extraction simulation to develop logistic regression (LR), random forest (RF), and decision tree (DT) algorithms that predict procedure step performance from operator, step, readability, and natural language processing-based features. Features were filtered using the Boruta approach. The algorithms were trained and optimized with a repeated 10-fold cross-validation. After training, inference was performed using variable importance and partial dependence plots. RESULTS: The RF, DT, and LR algorithms with all features had an area under the receiver operating characteristic curve (AUC) of 0.78, 0.77, and 0.75, respectively, and significantly outperformed the LR with only operator features (LROP), with an AUC of 0.61. The most important features were experience, familiarity, total words, and character-based metrics. The partial dependence plots showed that steps with fewer words, abbreviations, and characters were correlated with correct step performance. CONCLUSION: Machine learning algorithms are a promising approach for predicting step-level procedure performance, with acknowledged limitations on interpolating to nonobserved data, and may help guide procedure design after validation with additional data on further tasks. APPLICATION: After validation, the inferences from these models can be used to generate procedure design alternatives.

The Effects of Drowsiness Detection Technology and Education on Nurses' Beliefs and Attitudes toward Drowsy Driving.

Occupational ApplicationsNurses' perceived health threat from driving drowsy along with their attitude toward an intervention can be targeted to improve nurses' intentions to avoid this dangerous behavior. The evidence presented in this paper suggests that educational interventions that raise awareness of the risks of drowsy driving and its consequences (e.g., fatalities or injuries), as well as peer stories about their experiences, may positively affect nurses' perceived health threat and attitudes toward drowsy driving interventions.

Background Drowsy driving is prevalent among night-shift nurses, yet there is a gap in understanding nurses' beliefs and attitudes that may affect their intention to avoid drowsy driving.Objectives The objectives of the study were twofold: 1) investigate how behavioral constructs such as beliefs and attitudes may affect nurses' intention to avoid drowsy driving; and 2) assess changes in such beliefs and attitudes during a study that evaluated the effectiveness of educational and technological interventions.Methods Three-hundred night-shift nurses were recruited from a large hospital in Texas to participate in a randomized controlled trial. Participants were randomly assigned to three groups: 1) control; 2) educational intervention; and 3) combined educational and technological intervention. The study utilized an integrated model drawing from the constructs of the Theory of Planned Behavior and the Health Belief Model to elicit attitudes, beliefs, and intentions to use in-vehicle drowsiness detection technologies. Each group was surveyed pre- intervention and at post-intervention around 3 months later to assess changes in beliefs and attitudes. Structural equation models and path analysis were used to analyze changes in beliefs.Results Seventy-nine participants completed the pre-intervention questionnaire, and 44 nurses completed the pre- and post-intervention surveys. Intention was predicted primarily by attitude and perceived health threat. Perceived health threat also mediated the relationship between behavioral intention and the influence of subjective norms as well as perceived behavioral control. Participants who received education about drowsy driving had positive changes in beliefs.Conclusions Nurses' perceived health threat from driving drowsy and their attitude toward our intervention were important motivators to avoid drowsy driving. Interventions aiming at raising awareness of the risks associated with drowsy driving may be effective at motivating nurses to avoid drowsy driving.

Posttraumatic stress disorder hyperarousal event detection using smartwatch physiological and activity data.

Posttraumatic Stress Disorder (PTSD) is a psychiatric condition affecting nearly a quarter of the United States war veterans who return from war zones. Treatment for PTSD typically consists of a combination of in-session therapy and medication. However; patients often experience their most severe PTSD symptoms outside of therapy sessions. Mobile health applications may address this gap, but their effectiveness is limited by the current gap in continuous monitoring and detection capabilities enabling timely intervention. The goal of this article is to develop a novel method to detect hyperarousal events using physiological and activity-based machine learning algorithms. Physiological data including heart rate and body acceleration as well as self-reported hyperarousal events were collected using a tool developed for commercial off-the-shelf wearable devices from 99 United States veterans diagnosed with PTSD over several days. The data were used to develop four machine learning algorithms: Random Forest, Support Vector Machine, Logistic Regression and XGBoost. The XGBoost model had the best performance in detecting onset of PTSD symptoms with over 83% accuracy and an AUC of 0.70. Post-hoc SHapley Additive exPlanations (SHAP) additive explanation analysis showed that algorithm predictions were correlated with average heart rate, minimum heart rate and average body acceleration. Findings show promise in detecting onset of PTSD symptoms which could be the basis for developing remote and continuous monitoring systems for PTSD. Such systems may address a vital gap in just-in-time interventions for PTSD self-management outside of scheduled clinical appointments.

Understanding Heart Rate Reactions to Post-Traumatic Stress Disorder (PTSD) Among Veterans: A Naturalistic Study.

OBJECTIVE: We collected naturalistic heart rate data from veterans diagnosed with post-traumatic stress disorder (PTSD) to investigate the effects of various factors on heart rate. BACKGROUND: PTSD is prevalent among combat veterans in the United States. While a positive correlation between PTSD and heart rate has been documented, specific heart rate profiles during the onset of PTSD symptoms remain unknown. METHOD: Veterans were recruited during five cycling events in 2017 and 2018 to record resting and activity-related heart rate data using a wrist-worn device. The device also logged self-reported PTSD hyperarousal events. Regression analyses were performed on demographic and behavioral covariates including gender, exercise, antidepressants, smoking habits, sleep habits, average heart rate during reported hyperarousal events, age, glucocorticoids consumption, and alcohol consumption. Heart rate patterns during self-reported PTSD hyperarousal events were analyzed using Auto Regressive Integrated Moving Average (ARIMA). Heart rate data were also compared to an open-access non-PTSD representative case. RESULTS: Of 99 veterans with PTSD, 91 participants reported at least one hyperarousal event, with a total of 1023 events; demographic information was complete for 38 participants who formed the subset for regression analyses. The results show that factors including smoking, sleeping, gender, and medication significantly affect resting heart rate. Moreover, unique heart rate patterns associated with PTSD symptoms in terms of stationarity, autocorrelation, and fluctuation characteristics were identified. CONCLUSION: Our findings show distinguishable heart rate patterns and characteristics during PTSD hyperarousal events. APPLICATION: These findings show promise for future work to detect the onset of PTSD symptoms.

Bicyclist-evoked arousal and greater attention to bicyclists independently promote safer driving.

While attention has consistently been shown to be biased toward threatening objects in experimental settings, our understanding of how attention is modulated when the observer is in an anxious or aroused state and how this ultimately affects behavior is limited. In real-world environments, automobile drivers can sometimes carry negative perceptions toward bicyclists that share the road. It is unclear whether bicyclist encounters on a roadway lead to physiological changes and attentional biases that ultimately influence driving behavior. Here, we examined whether participants in a high-fidelity driving simulator exhibited an arousal response in the presence of a bicyclist and how this modulated eye movements and driving behavior. We hypothesized that bicyclists would evoke a robust arousal and orienting response, the strength of which would be associated with safer driving behavior. The results revealed that encountering a bicyclist evoked negative arousal by both self-report and physiological measures. Physiological and eye-tracking measures were themselves unrelated, however, being independently associated with safer driving behavior. Our findings offer a real-world demonstration of how arousal and attentional prioritization can lead to adaptive behavior.

The Impact of Commutes, Work Schedules, and Sleep on Near-Crashes during Nurses' Post Shift-Work Commutes: A Naturalistic Driving Study.

OCCUPATIONAL APPLICATIONSDriving and survey data were collected from nurses following the night-shift and analyzed with logistic regression and frequency analysis. The analyses showed that prior near-crashes and drive length contributed significantly to near-crashes. The frequency analysis showed that most near-crashes occurred on major roadways, including principal arterials, major collectors, and interstates, within the first 15 minutes of the drive. These results highlight the urgent need for countermeasures to prevent drowsy driving incidents among night-shift nurses. Specifically, nurses and hospital systems should focus on countermeasures that encourage taking a break on the post work commute and those that can intervene during the drive. This may include the use of educational programs to teach nurses the importance of adequate rest or taking a break to sleep during their drive home, or technology that can recognize drowsiness and alert nurses of their drowsiness levels, prompting them to take a break.

TECHNICAL ABSTRACTBackground Night-shift nurses are susceptible to drowsy driving crashes due to their long working hours, disrupted circadian rhythm, and reduced sleep hours. However, the extent to which work, sleep, and on-road factors impact the nurses' commutes and the occurrence of near-crash events is not well documented.Purpose A longitudinal naturalistic driving study with night-shift nurses from a large hospital in the United States was conducted to measure these factors and analyze the occurrence and location of near-crashes during post-shift commutes.Methods An on-board data recorder was used to record acceleration, speed, and GPS coordinates continuously. Nurses also completed daily surveys on their sleep, work, and commute. Near-crashes were identified from the data based on acceleration thresholds. Data from a total of 853 drives from 22 nurses and corresponding surveys were analyzed using Poisson and negative binomial regressions for swerve and hard brake near-crash events, respectively.Results Swerve events were increased by the length of the drive (RR = 2.59, LL = 1.62, UL = 4.16), and the occurrence of hard brakes (RR = 1.69, LL = 1.45, UL = 1.99), while hard brake events were increased by the occurrence of swerves (RR = 1.55, LL = 1.28, UL = 1.88). The majority of near-crashes occurred on principal arterials (n = 293), minor arterials (n = 71), and interstates (n = 51).Conclusions The results demonstrate the high risk of near-crashes during post-shift commutes, which may present danger to nurses and other drivers, and highlight the need for countermeasures that address shift structures, sleep quality, and taking breaks.

Steering or braking avoidance response in SHRP2 rear-end crashes and near-crashes: A decision tree approach.

OBJECTIVE: The paper presents a systematic analysis of drivers' crash avoidance response during crashes and near-crashes and developed a machine learning-based predictive model that can determine driver maneuver using pre-incident driver behavior and driving context. METHODS: We analyzed 286 naturalistic rear-end crashes and near-crashes from the SHRP2 naturalistic driving study. All the events were manually reduced using face video (face and forward) and kinematic responses. In this paper, we developed new reduction variables that enhanced the understanding of drivers' gaze behavior and roadway attention behavior during these events. These features reflected how the event criticality, measured using time to collision, related to drivers' pre-incident behavior (secondary behavior, gaze behavior), and drivers' perception of the event (physical reaction and maneuver). The imperative understanding of such relations was validated using a random forest- (RF) based classifier, which efficiently predicted if a driver was going to brake or change the lane as an avoidance maneuver. RESULTS: The RF presented in this paper effectively explored the nonlinear patterns in the data and was highly accurate (∼96 %) in its prediction. A further analysis of the RF model showed that six features played a pivotal role in the decision logic. These included the drivers' last glance duration before the event, last glance eccentricity, duration of 'eyes on road' immediately before the event, the time instance and criticality when the driver perceives the threat as well as acknowledge the threat, and possibility of an escape path in the adjacent lane. Using partial dependency plots, we also showed how different thresholds of these feature variables determined the drivers' maneuver intention. CONCLUSIONS: In this paper we analyzed driving context, drivers' behavior, event criticality, and drivers' response in a unified structure to predict their avoidance response. To the best of our knowledge, this is the first such effort where large-scale naturalistic data (crashes and near crashes) was analyzed for prediction of drivers' maneuver and determined key behavioral and contextual factors that contribute to this avoidance maneuver.

Toward a Taxonomy for Analyzing the Heart Rate as a Physiological Indicator of Posttraumatic Stress Disorder: Systematic Review and Development of a Framework.

BACKGROUND: Posttraumatic stress disorder (PTSD) is a prevalent psychiatric condition that is associated with symptoms such as hyperarousal and overreactions. Treatments for PTSD are limited to medications and in-session therapies. Assessing the way the heart responds to PTSD has shown promise in detecting and understanding the onset of symptoms. OBJECTIVE: This study aimed to extract statistical and mathematical approaches that researchers can use to analyze heart rate (HR) data to understand PTSD. METHODS: A scoping literature review was conducted to extract HR models. A total of 5 databases including Medical Literature Analysis and Retrieval System Online (Medline) OVID, Medline EBSCO, Cumulative Index to Nursing and Allied Health Literature (CINAHL) EBSCO, Excerpta Medica Database (Embase) Ovid, and Google Scholar were searched. Non-English language studies, as well as studies that did not analyze human data, were excluded. A total of 54 studies that met the inclusion criteria were included in this review. RESULTS: We identified 4 categories of models: descriptive time-independent output, descriptive and time-dependent output, predictive and time-independent output, and predictive and time-dependent output. Descriptive and time-independent output models include analysis of variance and first-order exponential; the descriptive time-dependent output model includes a classical time series analysis and mixed regression. Predictive time-independent output models include machine learning methods and analysis of the HR-based fluctuation-dissipation method. Finally, predictive time-dependent output models include the time-variant method and nonlinear dynamic modeling. CONCLUSIONS: All of the identified modeling categories have relevance in PTSD, although the modeling selection is dependent on the specific goals of the study. Descriptive models are well-founded for the inference of PTSD. However, there is a need for additional studies in this area that explore a broader set of predictive models and other factors (eg, activity level) that have not been analyzed with descriptive models.

Night-shift nurses and drowsy driving: A qualitative study.

BACKGROUND: Drowsy driving following the night shift is persistent among nurses resulting in elevated rates of vehicle crashes and crash-related injuries and deaths. While considerable effort has been dedicated to the development of countermeasures, implementation of these countermeasures in nursing has lagged behind other shift work oriented industries. Developing effective countermeasures for drowsy driving in nurses requires a thorough characterization of nurse's perceptions of drowsy driving and potential mitigations. OBJECTIVE: The objective of this research was to elicit night shift nurses' perceptions of drowsy driving, countermeasures, and educational and technological interventions. DESIGN: Perceptions were elicited through a semi-structured interview protocol. The protocol design was driven by previously identified research gaps. Questions focused on four topics: perceptions of drowsy driving, current practices and methods to mitigate drowsiness during the shift and commute, preferences and expectations for training on drowsiness management, and, preferences and expectations for technological mitigations. SETTING: The data collection took place at a large urban hospital in Texas, USA. PARTICIPANTS: Thirty night-shift nurses were recruited with voluntary sampling. No nurses declined to participate after initially consenting. The participants were male and female nurses who currently worked a 12 hour night shift. The nurses had between 1 and more than 20 years of experience and worked in a variety of units. METHOD: The interview recordings were transcribed by the research team and entered into a qualitative data analysis software. Transcripts were analyzed by two independent coders with a grounded theory approach to identify common themes and subthemes across participants. FINDINGS: Feelings of drowsiness typically manifested immediately following the shift or during the post work commute. Nurses responded to drowsiness by engaging in multiple ineffective countermeasures (e.g., listening to music) and effective countermeasures (e.g., naps) were used sparingly. Experiences and mitigation methods traversed through the nurses' social network although they did not always alter behavior. Nurses were uncertain but enthusiastic about educational and technological interventions preferring practical training and auditory interactive alerts. CONCLUSIONS: The findings suggest a strong need for real time drowsiness interventions during or immediately prior to nurses' post work commutes. Nurses' enthusiasm for training and technology to prevent drowsy driving suggests high levels of readiness and acceptance for such interventions. Future work should focus on the development and implementation of practical training and technological interventions for drowsy driving in nurses.

Classifying Major Depressive Disorder Using fNIRS During Motor Rehabilitation.

Major depressive disorder (MDD) has shown to negatively impact physical recovery in a variety of medical events (e.g., stroke and spinal cord injuries). Yet depression assessments, which are typically subjective in nature, are seldom considered to develop or guide rehabilitation strategies. The present study developed a predictive depression assessment technique using functional near-infrared spectroscopy (fNIRS) that can be rapidly integrated or performed concurrently with existing physical rehabilitation tasks. Thirty-one volunteers, including 14 adults clinically diagnosed with MDD and 17 healthy adults, participated in the study. Brain oxy-hemodynamic (HbO) responses were recorded using a 16-channel wearable continuous-wave fNIRS device while the volunteers performed the Grasp and Release Test in four 16-minute blocks. Ten features, extracted from HbO signals, from each channel served as inputs to XGBoost and Random Forest algorithms developed for each block and combination of successive blocks. Top 5 common features resulted in a classification accuracy of 92.6%, sensitivity of 84.8%, and specificity of 91.7% using the XGBoost classifier. This study identified mean HbO, full width half maximum and kurtosis, as specific neuromarkers, for predicting MDD across specific depression-related regions of interests (i.e., dorsolateral and ventrolateral prefrontal cortex). Our results suggest that a wearable fNIRS head probe monitoring specific brain regions, and limiting extraction to few features, can enable quick setup and rapid assessment of depression in patients. The overarching goal is to embed predictive neurotechnology during post-stroke and post-spinal-cord-injury rehabilitation sessions to monitor patients' depression symptomology so as to actively guide decisions about motor therapies.

Classification of Driver Distraction: A Comprehensive Analysis of Feature Generation, Machine Learning, and Input Measures.

OBJECTIVE: The objective of this study was to analyze a set of driver performance and physiological data using advanced machine learning approaches, including feature generation, to determine the best-performing algorithms for detecting driver distraction and predicting the source of distraction. BACKGROUND: Distracted driving is a causal factor in many vehicle crashes, often resulting in injuries and deaths. As mobile devices and in-vehicle information systems become more prevalent, the ability to detect and mitigate driver distraction becomes more important. METHOD: This study trained 21 algorithms to identify when drivers were distracted by secondary cognitive and texting tasks. The algorithms included physiological and driving behavioral input processed with a comprehensive feature generation package, Time Series Feature Extraction based on Scalable Hypothesis tests. RESULTS: Results showed that a Random Forest algorithm, trained using only driving behavior measures and excluding driver physiological data, was the highest-performing algorithm for accurately classifying driver distraction. The most important input measures identified were lane offset, speed, and steering, whereas the most important feature types were standard deviation, quantiles, and nonlinear transforms. CONCLUSION: This work suggests that distraction detection algorithms may be improved by considering ensemble machine learning algorithms that are trained with driving behavior measures and nonstandard features. In addition, the study presents several new indicators of distraction derived from speed and steering measures. APPLICATION: Future development of distraction mitigation systems should focus on driver behavior-based algorithms that use complex feature generation techniques.

Toward Computational Simulations of Behavior During Automated Driving Takeovers: A Review of the Empirical and Modeling Literatures.

OBJECTIVE: This article provides a review of empirical studies of automated vehicle takeovers and driver modeling to identify influential factors and their impacts on takeover performance and suggest driver models that can capture them. BACKGROUND: Significant safety issues remain in automated-to-manual transitions of vehicle control. Developing models and computer simulations of automated vehicle control transitions may help designers mitigate these issues, but only if accurate models are used. Selecting accurate models requires estimating the impact of factors that influence takeovers. METHOD: Articles describing automated vehicle takeovers or driver modeling research were identified through a systematic approach. Inclusion criteria were used to identify relevant studies and models of braking, steering, and the complete takeover process for further review. RESULTS: The reviewed studies on automated vehicle takeovers identified several factors that significantly influence takeover time and post-takeover control. Drivers were found to respond similarly between manual emergencies and automated takeovers, albeit with a delay. The findings suggest that existing braking and steering models for manual driving may be applicable to modeling automated vehicle takeovers. CONCLUSION: Time budget, repeated exposure to takeovers, silent failures, and handheld secondary tasks significantly influence takeover time. These factors in addition to takeover request modality, driving environment, non-handheld secondary tasks, level of automation, trust, fatigue, and alcohol significantly impact post-takeover control. Models that capture these effects through evidence accumulation were identified as promising directions for future work. APPLICATION: Stakeholders interested in driver behavior during automated vehicle takeovers may use this article to identify starting points for their work.

Using Topic Modeling to Develop Multi-level Descriptions of Naturalistic Driving Data from Drivers with and without Sleep Apnea.

One challenge in using naturalistic driving data is producing a holistic analysis of these highly variable datasets. Typical analyses focus on isolated events, such as large g-force accelerations indicating a possible near-crash. Examining isolated events is ill-suited for identifying patterns in continuous activities such as maintaining vehicle control. We present an alternative approach that converts driving data into a text representation and uses topic modeling to identify patterns across the dataset. This approach enables the discovery of non-linear patterns, reduces the dimensionality of the data, and captures subtle variations in driver behavior. In this study topic models are used to concisely described patterns in trips from drivers with and without untreated obstructive sleep apnea (OSA). The analysis included 5000 trips (50 trips from 100 drivers; 66 drivers with OSA; 34 comparison drivers). Trips were treated as documents, and speed and acceleration data from the trips were converted to "driving words." The identified patterns, called topics, were determined based on regularities in the co-occurrence of the driving words within the trips. This representation was used in random forest models to predict the driver condition (i.e., OSA or comparison) for each trip. Models with 10, 15 and 20 topics had better accuracy in predicting the driver condition, with a maximum AUC of 0.73 for a model with 20 topics. Trips from drivers with OSA were more likely to be defined by topics for smaller lateral accelerations at low speeds. The results demonstrate topic modeling as a useful tool for extracting meaningful information from naturalistic driving datasets.