Driver behavior and the use of automation in real-world driving.

BACKGROUND: The emergence of partial-automation in consumer vehicles is reshaping the driving task, the driver role, and subsequent driver behavior. When using partial-automation, drivers delegate the operational control of the dynamic driving task to the automation system, while remaining responsible for monitoring, object/event detection, response selection, and execution. Hence, driving has become a collaboration between driver and automation systems that is characterized by dynamic Transfers of Control (TOC). OBJECTIVE: This study aimed to assess how drivers leverage automation in real-world driving, identify driver and system-initiated TOCs, and provide a taxonomy to capture the underlying driver behaviors associated with automation disengagement. METHODS: Fourteen participants drove instrumented Cadillac CT6 vehicles for one-month each, yielding 1690 trips (22,108 miles), with a total of 5343 TOCs between manual driving, SAE Level 1 Adaptive Cruise Control (ACC), and SAE Level 2 Super Cruise (SC). RESULTS: The use of automation on limited access highways was prevalent (40 % of the miles driven were with SC and 10 % with ACC) yet not continuous. Drivers frequently initiated transitions between automation levels (mean = 9.98, SD = 8.32, transitions per trip), temporarily taking over the longitudinal and/or lateral vehicle control. These transitions were not necessarily related to immediate risk mitigation, but rather to the execution of functions beyond the automation system's capabilities or representing preferences in task execution. Driver-initiated TOCs from SC to manual driving followed the structure and temporal aspects of the hierarchical model of driver behavior. Strategic, Maneuver, and Control TOCs were associated with significantly different patterns of vehicle kinematics, automation disengagement modality, and TOC duration. System-initiated automation disengagements from SC to manual driving were rare (1%). CONCLUSIONS: Generalizing from objective, real-world driving data, this study provides an ecologically valid taxonomy for transfer of control building upon the hierarchical model of driver behavior. We show that driver-automation interactions can occur in each level of the hierarchical model and that TOCs are part of the driver's strategic, maneuver, and control levels of decision making. Thus, TOCs are not isolated or rare events, but rather an integral part of an ongoing, continuous and dynamic collaboration. This taxonomy contextualizes TOCs, paving the way for greater understanding of when and why drivers will takeover control, exposes the underlying motivations for TOCs, and characterizes how these are reflected in the driver's actions. The findings can inform the development of driver-centered automation systems as well as policies and guidelines for current and future automation levels.

Automated Driving System Collisions: Early Lessons.

OBJECTIVE: This research evaluated Automated Driving Systems (ADSs) involved collisions to identify factors relevant to future ADS research and development. BACKGROUND: Rapidly developing ADSs promise improved safety, among other benefits. Properly applied collision research can inform ADS development, to minimize future collisions. Errors and failures that result in collisions come from sources including the system, ADS operators, and external factors including other drivers. Partially automated systems incorporate new equipment and procedures creating new sources of human error. Fully autonomous systems represent a new class of drivers that interact in unique ways. METHOD: ADS collision reports from the California Department of Motor Vehicles and the National Transportation Safety Board were collected. An expert in human factors and collision investigation analyzed and categorized the crashes while extracting common factors. RESULTS: ADS vehicles were never at fault but were often affected from the rear during braking, turning, and gap acceptance maneuvers. Side impacts to ADS vehicles were related to passing vehicles and lane keeping behaviors. Unique incidents also provided additional insights. ADS collision rates cannot yet be determined with confidence. CONCLUSION: Conflicts that lead to collision-involvement with ADSs may be caused by differences between ADS and human driving behavior. Conservative ADS behavior may violate the expectations of other nearby human road users. APPLICATION: The findings from this work help inform the future development of ADS, as well as potentially the testing of ADS and the formation of policy to guide their future deployment.

Glass half-full: On-road glance metrics differentiate crashes from near-crashes in the 100-Car data.

BACKGROUND: Much of the driver distraction and inattention work to date has focused on concerns over drivers removing their eyes from the forward roadway to perform non-driving-related tasks, and its demonstrable link to safety consequences when these glances are timed at inopportune moments. This extensive literature has established, through the analyses of glance from naturalistic datasets, a clear relationship between eyes-off-road, lead vehicle closing kinematics, and near-crash/crash involvement. OBJECTIVE: This paper looks at the role of driver expectation in influencing drivers' decisions about when and for how long to remove their eyes from the forward roadway in an analysis that consider the combined role of on- and off-road glances. METHOD: Using glance data collected in the 100-Car Naturalistic Driving Study (NDS), near-crashes were examined separately from crashes to examine how momentary differences in glance allocation over the 25-s prior to a precipitating event can differentiate between these two distinct outcomes. Individual glance metrics of mean single glance duration (MSGD), total glance time (TGT), and glance count for off-road and on-road glance locations were analyzed. Output from the AttenD algorithm (Kircher and Ahlström, 2009) was also analyzed as a hybrid measure; in threading together on- and off-road glances over time, its output produces a pattern of glance behavior meaningful for examining attentional effects. RESULTS: Individual glance metrics calculated at the epoch-level and binned by 10-s units of time across the available epoch lengths revealed that drivers in near-crashes have significantly longer on-road glances, and look less frequently between on- and off- road locations in the moments preceding a precipitating event as compared to crashes. During on-road glances, drivers in near-crashes were found to more frequently sample peripheral regions of the roadway than drivers in crashes. Output from the AttenD algorithm affirmed the cumulative net benefit of longer on-road glances and of improved attention management between on- and off-road locations. CONCLUSION: The finding of longer on-road glances differentiating between safety-critical outcomes in the 100-Car NDS data underscores the importance of attention management in how drivers look both on and off the road. It is in the pattern of glances to and from the forward roadway that drivers obtained critical information necessary to inform their expectation of hazard potential to avoid a crash. APPLICATION: This work may have important implications for attention management in the context of the increasing prevalence of in-vehicle demands as well as of vehicle automation.

Conversation effects on neural mechanisms underlying reaction time to visual events while viewing a driving scene: fMRI analysis and asynchrony model.

This neuroimaging study investigated the neural mechanisms of the effect of conversation on visual event detection during a driving-like scenario. The static load paradigm, established as predictive of visual reaction time in on-road driving, measured reaction times to visual events while subjects watched a real-world driving video. Behavioral testing with twenty-eight healthy volunteers determined the reaction time effects from overt and covert conversation tasks in this paradigm. Overt and covert conversation gave rise to longer visual event reaction times in the surrogate driving paradigm compared to just driving with no conversation, with negligible effect on miss rates. The covert conversation task was then undertaken by ten right-handed healthy adults in a 4-Tesla fMRI magnet. We identified a frontal-parietal network that maintained event detection performance during the conversation task while watching the driving video. Increased brain activations for conversation vs. no conversation during such simulated driving was found not only in language regions (Broca's and Wernicke's areas), but also specific regions in bilateral inferior frontal gyrus, bilateral anterior insula and orbitofrontal cortex, bilateral lateral prefrontal cortex (right middle frontal gyrus and left frontal eye field), supplementary motor cortex, anterior and posterior cingulate gyrus, right superior parietal lobe, right intraparietal sulcus, right precuneus, and right cuneus. We propose an Asynchrony Model in which the frontal regions have a top-down influence on the synchrony of neural processes within the superior parietal lobe and extrastriate visual cortex that in turn modulate the reaction time to visual events during conversation while driving.

Semantic effects in naming perceptual identification but not in delayed naming: implications for models and tasks.

Previous research has demonstrated that the subjective danger and usefulness of words affect lexical decision times. Usually, an interaction is found: Increasing danger predicts faster reaction times (RTs) for words low on usefulness, but increasing danger predicts slower RTs for words high on usefulness. The authors show the same interaction with immediate auditory naming. The interaction disappeared with a delayed auditory naming control experiment, suggesting that it has a perceptual basis. In an attempt to separate input (signal to ear) from output (brain to muscle) processes in word recognition, the authors ran 2 auditory perceptual identification experiments. The interaction was again significant, but performance was best for words high on both danger and usefulness. This suggests that initial demonstrations of the interaction were reflecting an output approach/withdraw response conflict induced by stimuli that are both dangerous and useful. The interaction cannot be characterized as a tradeoff of speed versus accuracy.

Recognition of spoken words: semantic effects in lexical access.

Until recently most models of word recognition have assumed that semantic auditory naming effects come into play only after the identification of the word in question. What little evidence exists for early semantic effects in word recognition lexical decision has relied primarily on priming manipulations using the lexical decision task, and has used visual stimulus presentation. The current study uses semantics auditory stimulus presentation and multiple experimental tasks, and does not use priming. Response latencies for 100 common nouns were found to speech perception depend on perceptual dimensions identified by Osgood (1969): Evaluation, Potency, and Activity. In addition, the two-way interactions between these word recognition dimensions were significant. All effects were above and beyond the effects of concreteness, word length, frequency, onset phoneme characteristics, stress, and neighborhood density. Results are discussed against evidence from several areas of research suggesting a role of behaviorally important information in perception.