The Long and Winding Road: 25 Years of the National Advanced Driving Simulator.

The National Advanced Driving Simulator is a high-fidelity motion-base simulator owned by the National Highway Transportation Safety Administration and managed and operated by the University of Iowa. Its 25-year history has intersected with some of the most significant developments in automotive history, such as advanced driver assistance systems like stability control and collision warning systems, and highly automated vehicles. The simulator is an application of immersive virtual reality that uses multiprojection instead of head-mounted displays. A large-excursion motion system provides realistic acceleration and rotation cues to the driver. Due to its level of immersion and realism, drivers respond to events in the simulator the same way they would in their own vehicle. We document the history and technology behind this national facility.

Silent failure detection in partial automation as a function of visual attentiveness.

OBJECTIVE: Drivers using level 2 automation are able to disengage with the dynamic driving task, but must still monitor the roadway and environment and be ready to takeover on short notice. However, people are still willing to engage with non-driving related tasks, and the ways in which people manage this tradeoff are expected to vary depending on the operational design domain of the system and the nature of the task. Our aim is to model driver gaze behavior in level 2 partial driving automation when the driver is engaged in an email task on a cell phone. Both congested highway driving, traffic jams, and a hazard with a silent automation failure are considered in a driving simulator study conducted in the NADS-1 high-fidelity motion-based driving simulator. METHODS: Sequence analysis is a methodology that has grown up around social science research questions. It has developed into a powerful tool that supports intuitive visualizations, clustering analysis, covariate analyses, and Hidden Markov Models. These methods were used to create models for four different gaze behaviors and use the models to predict attention during the silent failure event. RESULTS: Predictive simulations were run with initial conditions that matched driver state just prior to the silent failure event. Actual gaze response times were observed to fall within distributions of predicted glances to the front. The three drivers with the largest glance response times were not able to take back manual control before colliding with the hazard. CONCLUSIONS: The simulated glance response time distributions can be used in more sophisticated ways when combined with other data. The glance response time probability may be conditioned on other variables like time on task, time of day, prevalence of the current behavior for this driver, or other variables. Given the flexibility of sequence analysis and the methods it supports (clustering, HMMs), future studies may benefit from its application to gaze behavior and driving performance data.

Multi-sensor driver monitoring for drowsiness prediction.

OBJECTIVE: Driver monitoring systems are growing in importance as well as capability. This paper reports drowsy driving detection models that use vehicular, behavioral, and physiological data. The objectives were to augment camera-based system with vehicle-based and heart rate variability measures from a wearable device and compare the performance of drowsiness detection models that use these data sources. Timeliness of the models in predicting drowsiness is analyzed. Timeliness refers to how quickly a model can identify drowsiness and, by extension, how far in advance of an adverse event a classification can be given. METHODS: Behavioral data were provided by a production-type Driver Monitoring System manufactured by Aisin Technical Center of America. Vehicular data were recorded from the National Advanced Driving Simulator's large-excursion motion-base driving simulator. Physiological data were collected from an Empatica E4 wristband. Forty participants drove the simulator for up to three hours after being awake for at least 16 hours. Periodic measurements of drowsiness were recorded every ten minutes using both observational rating of drowsiness by an external rater and the self-reported Karolinska Sleepiness Scale. Nine binary random forest models were created, using different combinations of data sources and ground truths. RESULTS: The classification accuracy of the nine models ranged from 0.77 to 0.92 on a scale from 0 to 1, with 1 indicating a perfect model. The best-performing model included physiological data and used a reduced dataset that eliminated missing data segments after heartrate variability measures were computed. The most timely model was able to detect the presence of drowsiness 6.7 minutes before a drowsy lane departure. CONCLUSIONS: The addition of physiological measures added a small amount of accuracy to the model performance. Models trained on observational ratings of drowsiness detected drowsiness earlier than those based only on Karolinska Sleepiness Scale, making them more timely in detecting the onset of drowsiness.

Benefits estimation of regenerative braking versus service braking.

A central question not yet examined in the literature is whether regenerative braking provides a kinematic deceleration safety advantage in time and distance over traditional service braking. This research explores three conditions of braking (traditional service braking, low level of regenerative braking, and high level of regenerative braking) to determine any safety advantages regenerative braking offers. Thirty participants took part in a simulator study with a between-subjects study design, allocating 10 participants per condition. The study drive took place in a simulator and involved three braking events. The results showed a significant difference between the means of the three conditions for average deceleration of the vehicle in the time interval between the driver releasing the accelerator and pressing the brake for all three events showing RB did provide the drivers with a braking advantage. When events 1 and 2 were combined, there was also significance with maximum brake force. Practitioner summary: This research looked to determine whether regenerative braking provides a deceleration safety advantage over traditional service braking. The results showed RB did provide the drivers with a braking advantage. The results also showed driver foot behaviour differed with the RB High condition.

Designing for the Extremes: Modeling Drivers' Response Time to Take Back Control From Automation Using Bayesian Quantile Regression.

OBJECTIVE: Understanding the factors that affect drivers' response time in takeover from automation can help guide the design of vehicle systems to aid drivers. Higher quantiles of the response time distribution might indicate a higher risk of an unsuccessful takeover. Therefore, assessments of these systems should consider upper quantiles rather than focusing on the central tendency. BACKGROUND: Drivers' responses to takeover requests can be assessed using the time it takes the driver to take over control. However, all the takeover timing studies that we could find focused on the mean response time. METHOD: A study using an advanced driving simulator evaluated the effect of takeover request timing, event type at the onset of a takeover, and visual demand on drivers' response time. A mixed effects model was fit to the data using Bayesian quantile regression. RESULTS: Takeover request timing, event type that precipitated the takeover, and the visual demand all affect driver response time. These factors affected the 85th percentile differently than the median. This was most evident in the revealed stopped vehicle event and conditions with a longer time budget and scenes with lower visual demand. CONCLUSION: Because the factors affect the quantiles of the distribution differently, a focus on the mean response can misrepresent actual system performance. The 85th percentile is an important performance metric because it reveals factors that contribute to delayed responses and potentially dangerous outcomes, and it also indicates how well the system accommodates differences between drivers.

The detection of drowsiness using a driver monitoring system.

Objective: Drowsiness is a major cause of driver impairment leading to crashes and fatalities. Research has established the ability to detect drowsiness with various kinds of sensors. We studied drowsy driving in a high-fidelity driving simulator and evaluated the ability of an automotive production-ready driver monitoring system (DMS) to detect drowsy driving. Additionally, this feature was compared to and combined with signals from vehicle-based sensors. Methods: The National Advanced Driving Simulator was used to expose drivers to long, monotonous drives. Twenty participants drove for about 4 h in the simulator between 10 p.m. and 2 a.m. They were allowed to use cruise control and traffic was sparse and semirandom, with both slower- and faster-moving vehicles. Observational ratings of drowsiness (ORDs) were used as the ground truth for drowsiness, and several dependent measures were calculated from vehicle and DMS signals. Drowsiness classification models were created that used only vehicle signals, only driver monitoring signals, and a combination of the 2 sources. Results: The model that used DMS signals performed better than the one that used only vehicle signals; however, the combination of the two performed the best. The models were effective at discriminating low levels of drowsiness from moderate to severe drowsiness; however, they were not effective at telling the difference between moderate and severe levels. A binary model that lumped drowsiness into 2 classes had an area under the receiver operating characteristic (ROC) curve of 0.897. Conclusions: Blinks and saccades have been shown to be predictive of microsleeps; however, it may be that detection of microsleeps and lane departures occurs too late. Therefore, it is encouraging that the model was able to distinguish mild from moderate drowsy driving. The use of automation may make vehicle-based signals useless for characterizing driver states, providing further motivation for a DMS. Future improvements in impairment detection systems may be expected through a combination of improved hardware, physiological measures from unobtrusive sensors and wearables, and the intelligent integration of environmental variables like time of day and time on task.

Gaze position modulates the effectiveness of forward collision warnings for drowsy drivers.

Advanced driver assistance systems (ADAS) have the potential to prevent crashes and reduce their severity. Forward collision warnings (FCW) are quickly becoming standard across vehicle lineups and may prevent frontal crashes by alerting drivers. Previous research has demonstrated the effectiveness of FCW for distracted drivers, but their effectiveness for other types of impairment remains unknown. Like distraction, drowsiness can impair driver response time and lead to crashes. The goal of the present study was to evaluate the effectiveness of FCW for moderately and severely drowsy drivers using a high-fidelity driving simulator. Drowsy drivers were divided into three warning conditions during a revealed stop vehicle forward collision event: An auditory alert, a haptic seat vibration, and a no warning baseline. Results indicate that FCW were effective at speeding drowsy driver response, but only when the drowsy drivers were looking away from the forward roadway at the onset of the event. These results have important implications for ADAS technology and driver state monitoring systems.

A contextual and temporal algorithm for driver drowsiness detection.

This study designs and evaluates a contextual and temporal algorithm for detecting drowsiness-related lane. The algorithm uses steering angle, pedal input, vehicle speed and acceleration as input. Speed and acceleration are used to develop a real-time measure of driving context. These measures are integrated with a Dynamic Bayesian Network that considers the time dependencies in transitions between drowsiness and awake states. The Dynamic Bayesian Network algorithm is validated with data collected from 72 participants driving the National Advanced Driving Simulator. The algorithm has a significantly lower false positive rate than PERCLOS-the current gold standard-and baseline, non-contextual, algorithms under design parameters that prioritize drowsiness detection. Under these parameters, the algorithm reduces false positive rate in highway and rural environments, which are typically problematic for vehicle-based detection algorithms. This algorithm is a promising new approach to driver impairment detection and suggests contextual factors should be considered in subsequent algorithm development processes. It may be combined with comprehensive mitigation methods to improve driving safety.

Evaluating driver drowsiness countermeasures.

OBJECTIVE: Driver drowsiness contributes to a substantial number of fatal and nonfatal crashes, with recent estimates attributing up to 21% of fatal crashes to drowsiness. This article describes recent NHTSA research on in-vehicle drowsiness countermeasures. Recent advances in technology and state detection algorithms have shown success in detecting drowsiness using a variety of data sources, including camera-based eye tracking, steering wheel position, yaw rate, and vehicle lane position. However, detection is just the first step in reducing drowsy driving crashes. Countermeasures are also needed to provide feedback to the driver, modify driver behavior, and prevent crashes. The goal of this study was to evaluate the effectiveness of in-vehicle drowsiness countermeasures in reducing drowsy lane departures. The tested countermeasures included different warning modalities in either a discrete or staged interface. METHODS: Data were collected from 72 young adult drivers (age 21-32) in the high-fidelity full-motion National Advanced Driving Simulator. Drivers completed a 45-min simulated nighttime drive at 2 time points, late night and early morning, where drowsiness was manipulated by continuous hours awake. Forty-eight drivers were exposed to one of 6 countermeasures that varied along 2 dimensions, type and modality. The countermeasures relied on a steering-based drowsiness detection algorithm developed in prior NHTSA research. Twenty-four drivers received no countermeasure and were used as a baseline comparison. System effectiveness was measured by lane departures and standard deviation in lateral position (SDLP). RESULTS: There was a reduction in drowsy lane departure frequency and lane position variability for drivers with countermeasures compared to the baseline no-countermeasure group. Importantly, the data suggest that multistage alerts, which provide an indication of increasing urgency, were more effective in reducing drowsy lane departures than single-stage discrete alerts, particularly during early morning drives when drivers were drowsier. CONCLUSIONS: The results indicate that simple in-vehicle countermeasures, such as an auditory-visual coffee cup icon, can reduce the frequency of drowsy lane departures in the context of relatively short drives. An important next step is to evaluate the impact of drowsiness countermeasures in the context of longer, multiple-hour drives. In these cases, it may not be possible to keep drivers awake via feedback warnings and it is important to understand whether countermeasures prompt drivers to stop to rest. The next phase of this research project will examine the role of drowsiness countermeasures over longer drives using a protocol that replicates the motivational conditions of drowsy driving.

Creating pedestrian crash scenarios in a driving simulator environment.

OBJECTIVE: In 2012 in the United States, pedestrian injuries accounted for 3.3% of all traffic injuries but, disproportionately, pedestrian fatalities accounted for roughly 14% of traffic-related deaths (NHTSA 2014 ). In many other countries, pedestrians make up more than 50% of those injured and killed in crashes. This research project examined driver response to crash-imminent situations involving pedestrians in a high-fidelity, full-motion driving simulator. This article presents a scenario development method and discusses experimental design and control issues in conducting pedestrian crash research in a simulation environment. Driving simulators offer a safe environment in which to test driver response and offer the advantage of having virtual pedestrian models that move realistically, unlike test track studies, which by nature must use pedestrian dummies on some moving track. METHODS: An analysis of pedestrian crash trajectories, speeds, roadside features, and pedestrian behavior was used to create 18 unique crash scenarios representative of the most frequent and most costly crash types. For the study reported here, we only considered scenarios where the car is traveling straight because these represent the majority of fatalities. We manipulated driver expectation of a pedestrian both by presenting intersection and mid-block crossing as well as by using features in the scene to direct the driver's visual attention toward or away from the crossing pedestrian. Three visual environments for the scenarios were used to provide a variety of roadside environments and speed: a 20-30 mph residential area, a 55 mph rural undivided highway, and a 40 mph urban area. RESULTS: Many variables of crash situations were considered in selecting and developing the scenarios, including vehicle and pedestrian movements; roadway and roadside features; environmental conditions; and characteristics of the pedestrian, driver, and vehicle. The driving simulator scenarios were subjected to iterative testing to adjust time to arrival triggers for the pedestrian actions. This article discusses the rationale behind creating the simulator scenarios and some of the procedural considerations for conducting this type of research. CONCLUSIONS: Crash analyses can be used to construct test scenarios for driver behavior evaluations using driving simulators. By considering trajectories, roadway, and environmental conditions of real-world crashes, representative virtual scenarios can serve as safe test beds for advanced driver assistance systems. The results of such research can be used to inform pedestrian crash avoidance/mitigation systems by identifying driver error, driver response time, and driver response choice (i.e., steering vs. braking).

Steering in a random forest: ensemble learning for detecting drowsiness-related lane departures.

OBJECTIVE: The aim of this study was to design and evaluate an algorithm for detecting drowsiness-related lane departures by applying a random forest classifier to steering wheel angle data. BACKGROUND: Although algorithms exist to detect and mitigate driver drowsiness, the high rate of false alarms and missed detection of drowsiness represent persistent challenges. Current algorithms use a variety of data sources, definitions of drowsiness, and machine learning approaches to detect drowsiness. METHOD: We develop a new approach for detecting drowsiness-related lane departures using steering wheel angle data that employ an ensemble definition of drowsiness and a random forest algorithm. Data collected from 72 participants driving the National Advanced Driving Simulator are used to train and evaluate the model. The model's performance was assessed relative to a commonly used algorithm, percentage eye closure (PERCLOS). RESULTS: The random forest steering algorithm had a higher classification accuracy and area under the receiver operating characteristic curve than PERCLOS and had comparable positive predictive value. The algorithm succeeds at identifying two key scenarios associated with the drowsiness detection task. These two scenarios consist of instances when drivers depart their lane because they fail to modulate their steering behavior according to the demands of the simulated road and instances when drivers correctly modulate their steering behavior according to the demands of the road. CONCLUSION: The random forest steering algorithm is a promising approach to detect driver drowsiness. The algorithm's ties to consequences of drowsy driving suggest that it can be easily paired with mitigation systems.

Effects of T2-Weighted MRI Based Cranial Volume Measurements on Studies of the Aging Brain.

Many brain aging studies use total intracranial volume (TIV) as a proxy measure of premorbid brain size that is unaffected by neurodegeneration. T1-weighted Magnetic Resonance Imaging (MRI) sequences are commonly used to measure TIV, but T2-weighted MRI sequences provide superior contrast between the cerebrospinal fluid (CSF) bounding the premorbid brain space and surrounding dura mater. In this study, we compared T1-based and T2-based TIV measurements to assess the practical impact of this superior contrast on studies of brain aging. 810 Alzheimer's Disease Neuroimaging Initiative (ADNI) participants, including healthy elders and those with mild cognitive impairment (MCI) and Alzheimer's Disease (AD), received T1-weighted and T2-weighted MRI at their baseline evaluation. TIV was automatically estimated from T1-weighted images using FreeSurfer version 4.3 (T1TIV), and an automated active contour method was used to estimate TIV from T2-weighted images (T2TIV). The correlation between T1TIV and T2TIV was high (.93), and disagreement was greater on larger heads. However, correcting a FreeSurfer-based measure of total parenchymal volume by dividing it by T2TIV led to stronger expected associations with a standardized measure of cognitive dysfunction (MMSE) in Poisson regression models among individuals with AD (z=1.73 vs. 1.09) and MCI (z=3.15 vs. 2.79) than a corresponding parenchymal volume measure divided by T1TIV. This effect was enhanced when the analysis was restricted to the cases where T1TIV and T2TIV disagreed the most. These findings suggest that T2-based TIV measurements may be higher fidelity than T1-based TIV measurements, thus leading to greater sensitivity to detect biologically plausible brain-behavior associations.