Trust is essential for effective human–AI teaming, yet AI agents can violate trust through both performance failures and misaligned intentions. While prior work has emphasized performance-based errors, it remains unclear how violations of cooperative intent, purpose-based violations, impact trust and how trust can be repaired. To address this, we designed a game-theoretic experiment manipulating both performance and purpose violations by an AI teammate pursuing a shared goal. We also tested three trust repair strategies: no response, apology with explanation, and apology with promise. Results showed that purpose-based violations elicited significantly greater trust drop than performance-based ones. Moreover, an apology with explanation was effective in repairing trust after purpose-based violations. These findings underscore the importance of distinguishing between types of trust violations and tailoring repair strategies accordingly. Designers of AI teammates should consider integrating informative explanations, especially when addressing goals misalignment, to restore trust and support long-term collaboration.

This paper presents a stochastic calibration method based on Approximate Bayesian Computation (ABC). This method is applied to calibrate two car-following control models: linear control and model predictive control (MPC). The method is likelihood-function-free, where the likelihood function is replaced by simulation to approximate the posterior distribution of model parameters. This structure affords flexibility to calibrate posterior joint distributions of complex models, even those without analytical closed forms such as MPC. Two experiments were conducted to evaluate how well the proposed method reproduces: (i) marginal and joint distributions of model parameters, using synthetic data and (ii) vehicle trajectories (acceleration, speed, and position), using field data involving two commercial adaptive cruise control (ACC) systems. The results showed that the ABC method can reproduce marginal and joint distributions reasonably well for the linear controller as well as the non-analytical MPC-based controller, which was previously infeasible. The method can also robustly characterize the commercial ACC behavior at the trajectory level, which suggests that the simple linear controller better describes their behavior.

In hybrid mobility societies, where automated vehicles (AVs) and humans interact in public spaces, the significance of prosocial behaviors intensifies. These behaviors are crucial for the smooth functioning of an interdependent transportation environment, mitigating challenges from the integration of AVs and human-operated systems, and enhancing user well-being by fostering more efficient, less stressful, and inclusive environments. This study explores the impact of receiving prosocial behaviors on cognition, riding behavior, and well-being of micromobility users through interdependent traffic situations within a simulated urban environment. Our mixed design study involved two types of social interactions as between-subject conditions of prosocial and asocial interaction, and three categories of time constraint as within-subject conditions: relaxed, neutral, and pressed. The findings reveal that receiving prosocial and asocial behaviors can affect the state of well-being and trial performance in a mobility environment.

Human-AI conversation provides a natural, unobtrusive, yet under-explored way to investigate trust dynamics in human-AI teams (HATs). In this paper, we modeled dynamic trust evolution in conversations using a novel method, trajectory epistemic network analysis (T-ENA). T-ENA captures the multidimensional aspect of trust (i.e., analytic and affective), and trajectory analysis segments conversations to capture temporal changes of trust over time. Twenty-four participants performed a habitat maintenance task assisted by a conversational agent and verbalized their experiences and feelings after each task. T-ENA showed that agent reliability significantly affected people’s conversations in the analytic process of trust, t(38.88)=15.18,p<0.001,Cohen′s d=4.72, such as discussing agents’ errors. The trajectory analysis showed that trust dynamics manifested through conversation topic diversity and flow. These results showed trust dimensions and dynamics in conversation should be considered interdependently and suggested that an adaptive conversational strategy for managing trust in HATs.

Self-driving vehicles promise many safety, mobility, and environmental benefits. However, users’ lack of trust and acceptance may threaten the success and potential of this technology. Monitoring the driver’s emotional state is one way to address this challenge. Empathetic automation can respond to the driver’s state and improve the experience and acceptance of self-driving vehicle drivers. In this study, 24 participants rode in a self-driving vehicle simulator and experienced three automation styles (aggressive, moderate, conservative) and four intersection types (with and without a stop sign, and with and without traffic.) We identified the observed drivers’ emotions from the video data and labeled the video frames using the dimensional and discrete emotion models to examine how automation behavior affects the driver’s emotional state. We used multilevel Bayesian linear regression and multilevel Dirichlet regression to model the continuous and discrete emotions, respectively. The automation driving style effect varied for each participant. The same conditions provoked positive responses for some participants, and negative for others. Furthermore, the results showed that intersection type, the position within the intersection, and their interaction affected the driver’s emotional state. This indicates that personalized driver state monitoring systems might enhance drivers’ experience in self-driving vehicles.

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.

Objective This study evaluates the benefits and costs associated with providing drivers continuous feedback on the limits and behavior of imperfect vehicle control automation. Background In-vehicle automated systems remove drivers from active vehicle control, often at the expense of timely interventions when failures occur. Discrete warnings, as a type of feedback to inform drivers about automated system behavior, fail to keep drivers aware of its proximity to operating limits. Method In a fixed-based simulator, 48 drivers drove using Adaptive Cruise Control (ACC)—a form of control automation that maintains a set speed, or a set headway if the vehicle encounters a slower moving vehicle. A first experiment compared ACC with discrete warnings to ACC with continuous information, which indicated moment-to-moment ACC state relative to its operating limits. Three display conditions, designed to provide non-obtrusive, ecologically-valid information, were evaluated in a second experiment: 1) a visual interface; 2) an auditory interface; and 3) a combined visual-auditory interface. Results Drivers provided with continuous feedback relied more appropriately on ACC than did those with discrete warnings. Continuous feedback increased the frequency of proactive responses to automation failures and improved system understanding. Of the three displays, the combined visual-auditory interface performed the best. Conclusion Continuous feedback helped communicate to drivers the evolving relationship between system performance and operating limits. Application Displays for increasingly automated vehicles should inform about the automation's situation-specific behavior rather than simply alert drivers to failures and/or the need to resume vehicle control in order to promote appropriate understanding and trust.

For older adults, familiarity plays an important role in reducing driving errors, improving wayfinding, and maintaining driving confidence. However, only a few studies have explored the influence of familiarity on driving behavior, route risk, and route choice among older drivers. The vehicles of 29 drivers age 65 and older were instrumented with on-board diagnostic devices for four months to record their routes driven, and risky driving behavior events. For each route driven, trip diaries were used to provide participants with retrospective feedback of their risky driving behaviors, alternate low-risk routes, and record answers to three questions pertaining to: familiarity with the route driven, suggested low-risk route alternative, and reasons for choosing a route. When familiarity responses were regressed on the number of risky driving behavior events, results showed that for every 10% increase in familiarity with driven routes, a 0.07 increase in risky driving behavior incidents per trip was estimated. Results from the generalized mixed effects regression showed that older drivers were less likely to drive a suggested low-risk route when they were more familiar with the alternate routes. Content analysis conducted on the reasons for choosing a familiar route showed that older drivers preferred routes that were direct, had less traffic, and depended on the trip purpose, such as running multiple errands. These results provide important insights on the influence of familiarity on route choice, preferences, and driving behaviors of older drivers and provide opportunities for developing targeted route choice models for navigation systems that can include factors such as familiarity.

Performance-cost trade-offs in video object tracking tasks for long video sequences is investigated. A novel frame-subsampled, drift-resilient (FSDR) video object tracking algorithm is presented that would achieve desired tracking accuracy while dramatically reducing computing time by processing only sub-sampled video frames. A new pattern matching score metric is proposed to estimate the probability of drifting. A drift-recovery procedure is developed to enable the algorithm to recover from a drift situation and resume accurate tracking. Compared against state-of-the-art video object tracking algorithms, dramatic performance (accuracy) enhancement and cost (computing time) reduction are observed.

A novel frame-subsampled, drift-resilient (FSDR) video object tracking (VOT) algorithm is proposed. Two design goals are sought: to improve the accuracy and to reduce the processing time. The drifting problem is mitigated with a drift detector and accompanying drift recovery mechanism. When a drift is detected, the recovery mechanism provides an opportunity to put the tracking back on track. To gather context-dependent statistics required for these procedures, an initial short segment of the video sequence will be used as a training sequence. Thus, this algorithm is more suitable for video sequences much longer than several minutes. To reduce computing time, a novel frame-subsampling strategy is proposed to process the VOT on small subset of frames. The trajectory of the tracked object on frames that are skipped will be estimated via interpolation. Compared with state of art VOT algorithms, dramatic improvement of performance (accuracy) and orders of magnitude computing time reduction are observed.

Older drivers face decline in perceptual, cognitive, and motor abilities, and yet, increased fragility largely explains their increased risk of fatal crashes. Adaptation and self-regulation explain why older drivers can be safe drivers in the face of declining ability. Left turns and U-turns are particularly challenging, accounting for 10% of crashes for drivers aged 60-69 and increasing to 32% for those over 80. To mitigate these driving challenges through more effective adaptation, a route risk measure was developed. The route risk measure quantifies the risk of driving challenges such as left turns, U-turns, and trip length using older driver crash statistics from the General Estimates System. We applied the risk measure to 1253 trips taken by 39 urban and rural older drivers residing in Wisconsin. A search for a low-risk alternative route was conducted by applying the measure to corresponding routes suggested by Google Maps. Results showed that the low-risk alternative reduced risk for 77.7% of the older drivers' trips, on average, by 61.4%. The low-risk alternative had 1.50 fewer left turns and 0.23 fewer U-turns and were 0.44 mi shorter. Thus, selecting low-risk alternatives from the routes suggested by Google could help drivers avoid challenging maneuvers, potentially reducing their crash risk by promoting more effective adaptation to their declining abilities.

Human-technology interactions involving errors undermine acceptance and performance. The effect of errors and the ability to recover from them represent a particularly important consideration for design in safety-critical multitasking situations. However, few studies have considered the recovery process of errors in multitasking situations, such as their contribution to driver distraction. This paper investigates errors that drivers make interacting with an infotainment system. In this study, participants (N = 46) drove a stimulated vehicle and performed word entry tasks on a touch screen. Errors undermined driving and task performance. We also identified four different error recovery strategies and found that the accumulated information related to the driving situation and the characteristics of an infotainment system affected the choice of strategy. Implications for in-vehicle interface design, driver models, and general multitasking design are discussed.

Drivers show a wide range of behavior while performing a secondary task behind the wheel. In the current study, we categorized drivers into groups based on their glance behavior at task boundary (i.e., pressing a touch screen button after reading driving-related messages) and compared driving capabilities of drivers in each group. The comparison between the groups identifies different eye glance strategies, or task switching decisions, and associated vehicle control behaviors. Senders' uncertainty model was adapted to explain the results and to suggest future directions in developing driver models.

Driver distraction represents an increasingly important contributor to crashes and fatalities. Technology that can detect and mitigate distraction by alerting distracted drivers could play a central role in maintaining safety. Based on either eye measures or driver performance measures, numerous algorithms to detect distraction have been developed. Combining both eye glance and vehicle data could enhance distraction detection. The goal of this paper is to evaluate whether changes in the eye-steering correlation structure can indicate distraction. Drivers performed visual, cognitive, and cognitive/visual tasks while driving in a simulator. The auto- and cross-correlations of horizontal eye position and steering wheel angle show that eye movements associated with road scanning produce a low eye-steering correlation. However, even this weak correlation is sensitive to distraction. Time lead associated with the maximum correlation is sensitive to all three types of distraction, and the maximum correlation coefficient is most strongly affected by off-road glances. These results demonstrate that eye-steering correlation statistics can detect distraction and differentiate between types of distraction.

Connected vehicles offer great potential for new sources of information, but may also introduce new sources of distraction. This paper compares three methods to quantify distraction, and focuses on one method: computational models of driver behavior. An integration of a saliency map and the Distract-R prototyping and evaluation system is proposed as a potential model. The saliency map captures the bottom-up influences of visual attention and this influence is integrated with top-down influences captured by Distract-R. The combined model will assess the effect of coordinating salient visual features and drivers' expectations, and in using both together, generate more robust predictions of performance.

Detection of alcohol-induced driving impairment through vehicle-based sensor signals is of paramount importance for road safety. To differentiate the driving conditions with and without alcohol-induced impairment, data were collected from 108 drivers under both conditions in a high-fidelity driving simulator. With this data set, various quantitative measures of steering wheel movement, including not only simple statistics such as the mean and the standard deviation but nonlinear dynamic invariant measures such as sample entropy and Lyapunov exponent as well, are compared in terms of their differentiating capabilities. Nonlinear invariant measures are more robust and consistent than the simple measures in differentiating the impairment. Furthermore, people respond to alcohol-induced impairment quite differently, and for a certain group of people, the alcohol-induced impairment can be well detected using these nonlinear invariant measures. Many interesting insights into characterizing the effect of alcohol on driving behavior are obtained in this paper. This paper lays a foundation for the future development of a real-time detection method for alcohol-induced impairment.

In-vehicle information systems (IVISs) can enhance or compromise driving safety. Such systems present an array of messages that range from collision warnings and navigation instructions to tire pressure and e-mail alerts. If these messages are not properly managed, the IVIS might fail to provide the driver with critical information, which could undermine safety. In addition, if the IVIS simultaneously presents multiple messages, the driver may fail to attend to the most critical information. To date, only simple algorithms that use priority-based filters have been developed to address this problem. This paper presents a dynamic programming model that goes beyond the immediate relevance and urgency parameters of the current Society of Automotive Engineers (SAE) message scheduling algorithm. The resulting algorithm considers the variation of message value over time, which extends the planning horizon and creates a more valuable stream of messages than that based only on the instantaneous message priority. This method has the potential to improve road safety because the most relevant information is displayed to drivers across time and not just the highest priority at any given instant. Applying this algorithm to message sets shows that scheduling that considers the time-based message value, in addition to priority, results in substantially different and potentially better message sequences compared with those based only on message priority. This method can be extended to manage driver workload by adjusting message timing relative to demanding driving maneuvers.

As use of in-vehicle information systems (IVISs) such as cell phones, navigation systems, and satellite radios has increased, driver distraction has become an important and growing safety concern. A promising way to overcome this problem is to detect driver distraction and adapt in-vehicle systems accordingly to mitigate such distractions. To realize this strategy, this paper applied support vector machines (SVMs), which is a data mining method, to develop a real-time approach for detecting cognitive distraction using drivers' eye movements and driving performance data. Data were collected in a simulator experiment in which ten participants interacted with an IVIS while driving. The data were used to train and test both SVM and logistic regression models, and three different model characteristics were investigated: how distraction was defined, which data were input to the model, and how the input data were summarized. The results show that the SVM models were able to detect driver distraction with an average accuracy of 81.1%, outperforming more traditional logistic regression models. The best performing model (96.1% accuracy) resulted when distraction was defined using experimental conditions (i.e., IVIS drive or baseline drive), the input data were comprised of eye movement and driving measures, and these data were summarized over a 40-s window with 95% overlap of windows. These results demonstrate that eye movements and simple measures of driving performance can be used to detect driver distraction in real time. Potential applications of this paper include the design of adaptive in-vehicle systems and the evaluation of driver distraction

Appropriate trust in and reliance on automation are critical for safe and efficient system operation. This paper fills an important research gap by describing a quantitative model of trust in automation. We extend decision field theory (DFT) to describe the multiple sequential decisions that characterize reliance on automation in supervisory control situations. Extended DFT (EDFT) represents an iterated decision process and the evolution of operator preference for automatic and manual control. The EDFT model predicts trust and reliance, and describes the dynamic interaction between operator and automation in a closed-loop fashion: the products of earlier decisions can transform the nature of later events and decisions. The simulation results show that the EDFT model captures several consistent empirical findings, such as the inertia of trust and the nonlinear characteristics of trust and reliance. The model also demonstrates the effects of different types of automation on ttrust and reliance. It is possible to expand the EDFT model for multioperator multiautomation situations.

As computers and other information technology move into cars and trucks, distraction-related crashes are likely to become an important problem. This paper begins to address this problem by examining how alert strategy (graded and single-stage) and alert modality (haptic and auditory) affect how well collision warning systems mitigate distraction and direct drivers attention to the car ahead when it unexpectedly brakes. We conducted two experiments in which drivers interacted with an in-vehicle email system and a collision warning system signaled a braking lead vehicle. The first experiment showed that graded alerts led to a greater safety margin and a lower rate of inappropriate responses to nuisance warnings. A second experiment focused on attitudes toward the collision warning system and found that graded alerts were more trusted than single stage alerts and that haptic alerts, a vibrating seat in these experiments, were perceived as less annoying and more appropriate. Graded haptic alerts offer a promising approach to developing context aware computing in a safety-critical application.

The increasing technological sophistication of ship navigation systems may significantly alter the skills, knowledge, and strategies involved in navigating large ships. Many examples in other domains illustrate the dangers of technology-driven innovations. These examples show that without a systematic method to detect design flaws and training requirements, technology-driven designs may degrade rather than enhance maritime safety. The operator function model (OFM) provides the basis for examining technological innovations; however, the OFM does not describe specific cognitive demands. Augmenting the OFM with a description of cognitive operations provides a structured cognitive task analysis tool-OFM-COG-that can identify the design and training requirements needed to safeguard system performance. This approach identifies how to tailor designs, develop training, and adjust qualifications to minimize the human errors that might otherwise accompany technological innovation. The paper shows how OFM-COG can catalog differences between traditional navigation systems and those augmented with electronic charts and collision avoidance systems. Specifically, it examines the cognitive demands of collision avoidance and track keeping, with and without advanced technological aids. This analysis demonstrates that some advanced radars may in fact increase the likelihood of certain collisions, and that the current certification process does not reflect the cognitive demands of the new technology. The analysis also indicates that electronic chart display and information systems (ECDIS) can reduce the redundancy that has served to make traditional systems quite reliable. Drawing upon these examples, the paper describes OFM-COG and demonstrates how this model-based analysis technique can document the cognitive implications of technological innovations.

A study was conducted to determine what factors affect how humans detect correlation when forming mental models of complex systems. Using a paradigm that elicits the subject's mental model of a system by the subject's reconfiguration of the display, the authors conducted four experiments that illustrate humans' ability and factors affecting their ability to detect correlation in a dynamic display. It is shown that subjects can detect correlation in a dynamic system and that lowering the correlation makes detecting the correlation progressively more difficult. The data, combined with the subjects' comments, shows that the strategies adopted by the subjects govern their ability to detect correlation. The strategies adopted depend on the degree of correlation and the type of movement. Correlation predisposed subjects to discover grouping by either pairwise comparison (in the case of low correlation) or global search (in the case of high correlation). The type of movement, on the other hand, predisposed subjects to group bars by height (in the case of random walk movement) or correlation (in the case of Gaussian mean zero movement). Generally, this study forms the beginning of a research program investigating the factors that guide operators in generating mental models of complex systems.>

Summary form only given. The importance of mental models in complex control tasks is discussed, and the lattice theory of mental models is described. Plans for a series of experiments using a simulation developed with the Lab View programming language to validate the lattice theory of mental models are reported.>