In highly automated driving, drivers occasionally need to take over control of the car due to limitations of the automated driving system. Research has shown that visually distracted drivers need about 7 s to regain situation awareness (SA). However, it is unknown whether the presence of a hazard affects SA. In the present experiment, 32 participants watched animated video clips from a driver's perspective while their eyes were recorded using eye-tracking equipment. The videos had lengths between 1 and 20 s and contained either no hazard or an impending crash in the form of a stationary car in the ego lane. After each video, participants had to (1) decide (no need to take over, evade left, evade right, brake only), (2) rate the danger of the situation, (3) rebuild the situation from a top-down perspective, and (4) rate the difficulty of the rebuilding task. The results showed that the hazard situations were experienced as more dangerous than the non-hazard situations, as inferred from self-reported danger and pupil diameter. However, there were no major differences in SA: hazard and non-hazard situations yielded equivalent speed and distance errors in the rebuilding task and equivalent self-reported difficulty scores. An exception occurred for the shortest time budget (1 s) videos, where participants showed impaired SA in the hazard condition, presumably because the threat inhibited participants from looking into the rear-view mirror. Correlations between measures of SA and decision-making accuracy were low to moderate. It is concluded that hazards do not substantially affect the global awareness of the traffic situation, except for short time budgets.

In conditionally automated driving, drivers do not have to monitor the road, whereas in partially automated driving, drivers have to monitor the road permanently. We evaluated a dynamic allocation of monitoring tasks to human and automation by providing a monitoring request (MR) before a possible take-over request (TOR), with the aim to better prepare drivers to take over safely and efficiently. In a simulator-based study, an MR + TOR condition was compared with a TOR-only condition using a within-subject design with 41 participants. In the MR + TOR condition, an MR was triggered 12 s before a zebra crossing, and a TOR was provided 7 s after the MR onset if pedestrians crossing the road were detected. In the TOR-only condition, a TOR was provided 5 s before the vehicle would collide with a pedestrian if the participant did not intervene. Participants were instructed to perform a self-paced visual-motor non-driving task during automated driving. Eye tracking results showed that participants in the MR + TOR condition responded to the MR by looking at the driving environment. They also exhibited better take-over performance, with a shorter response time to the TOR and a longer minimum time to collision as compared to the TOR-only condition. Subjective evaluations also showed advantages of the MR: participants reported lower workload, higher acceptance, and higher trust in the MR + TOR condition as compared to the TOR-only condition. Participants’ reliance on automation was tested in a third drive (MR-only condition), where automation failed to provide a TOR after an MR. The MR-only condition resulted in later responses (and errors of omission) as compared to the MR + TOR condition. It is concluded that MRs have the potential to increase safety and acceptance of automated driving as compared to systems that provide only TORs. Drivers’ trust calibration and reliance on automation need further investigation.

To solve the problem that existing driving data cannot correlate to the large number of vehicles in terms of driving risks, is the functionality of intelligent driving algorithm should be improved. This paper deeply explores driving data to build a link between massive driving data and a large number of sample vehicles for driving risk analysis. It sorted out certain driving behavior parameters in the driving data, and extracted some parameters closely related to the driving risk; it further utilized the principal component analysis and factor analysis in spatio-temporal data to integrate certain extracted parameters into factors that are clearly related to the specific driving risks; then, it selected factor scores of driving behaviors as indexes for hierarchical clustering, and obtained multi-level clustering results of the driving risks of corresponding vehicles; in the end, it interpreted the clustering results of the vehicle driving risks. According to the results, it is found that cluster for different risks proposed in this paper for driving behaviors is effective in the hierarchical cluster for typical driving behaviors and it also offers a solution for risk analyses between driving data and large sample vehicles. The results provide the basis for training on safe driving for the key vehicles, and the improvement of advanced driver assistance system, which shows a wide application prospect in the field of intelligent drive.

Although extensive research has been conducted to design path-following algorithms for automated vehicles, the cross comparison between different path-following controllers is still weakly-analyzed. Therefore, we benchmarked five path-following algorithms to evaluate their performance according to various disturbances like gust wind, drop of road friction coefficient and inaccurate GPS localization. The comparison was carried out in simulation environment between geometrical-based, path controller with preview, LQR, linear MPC and observer-based controller with integral action approaches.

Across the automotive industry, manufacturers have recently released various Partial Automation systems (SAE Level 2) which allow simultaneous/combined execution of both lateral and longitudinal vehicle control at the same time, yet still require active human supervision/engagement. Current reactive trends will be reviewed across major automotive players regarding differences in terminology, HMI input/outputs, and escalation intervals. Scholarly research is also reviewed pertaining to proactive strategies for driver engagement. Additionally, human factors research and findings will be presented regarding recommendations for situation awareness, human machine interfaces, TOR, as well as shared control concepts. The tutorial will conclude with discussion and brainstorming around outlook toward tele-operated remote driving services (Tele-Driving); what they have to offer beyond assisted/automated driving, autonomous vehicles, and ride-hailing/car-sharing paradigms; as well as the design/conduct of human factors research regarding Tele-Driving.

This paper aims to recognize driving risks in individual vehicles online based on a data-driven methodology. Existing advanced driver assistance systems (ADAS) have difficulties in effectively processing multi-source heterogeneous driving data. Furthermore, parameters adopted for evaluating the driving risk are limited in these systems. The approach of data-driven modeling is investigated in this study for utilizing the accumulation of on-road driving data. A recognition model of driving risk based on belief rule-base (BRB) methodology is built, predicting driving safety as a function of driver characteristics, vehicle state and road environment conditions. The BRB model was calibrated and validated using on-road data from 30 drivers. The test results show that the recognition accuracy of our proposed model can reach about 90% in all situations with three levels (none, medium, large) of driving risks. Furthermore, the proposed simplified model, which provides real-time operation, is implemented in a vehicle driving simulator as a reference for future ADAS and belongs to research on artificial intelligence (AI) in the automotive field.

A common challenge with processing naturalistic driving data is that humans may need to categorize great volumes of recorded visual information. By means of the online platform CrowdFlower, we investigated the potential of crowdsourcing to categorize driving scene features (i.e., presence of other road users, straight road segments, etc.) at greater scale than a single person or a small team of researchers would be capable of. In total, 200 workers from 46 different countries participated in 1.5. days. Validity and reliability were examined, both with and without embedding researcher generated control questions via the CrowdFlower mechanism known as Gold Test Questions (GTQs).By employing GTQs, we found significantly more valid (accurate) and reliable (consistent) identification of driving scene items from external workers. Specifically, at a small scale CrowdFlower Job of 48 three-second video segments, an accuracy (i.e., relative to the ratings of a confederate researcher) of 91% on items was found with GTQs compared to 78% without. A difference in bias was found, where without GTQs, external workers returned more false positives than with GTQs. At a larger scale CrowdFlower Job making exclusive use of GTQs, 12,862 three-second video segments were released for annotation. Infeasible (and self-defeating) to check the accuracy of each at this scale, a random subset of 1012 categorizations was validated and returned similar levels of accuracy (95%).In the small scale Job, where full video segments were repeated in triplicate, the percentage of unanimous agreement on the items was found significantly more consistent when using GTQs (90%) than without them (65%). Additionally, in the larger scale Job (where a single second of a video segment was overlapped by ratings of three sequentially neighboring segments), a mean unanimity of 94% was obtained with validated-as-correct ratings and 91% with non-validated ratings. Because the video segments overlapped in full for the small scale Job, and in part for the larger scale Job, it should be noted that such reliability reported here may not be directly comparable. Nonetheless, such results are both indicative of high levels of obtained rating reliability.Overall, our results provide compelling evidence for CrowdFlower, via use of GTQs, being able to yield more accurate and consistent crowdsourced categorizations of naturalistic driving scene contents than when used without such a control mechanism. Such annotations in such short periods of time present a potentially powerful resource in driving research and driving automation development.

Vehicle field test can be conducted smoothly because of the automobile-mounted monitoring system and abundant diving data have been collected. Driving data mining is in an urgent need of new thoughts introduced to break through the original technical bottleneck. This paper presented a novel method of symbolic representation in diving data mining and applied the idea of time series symbolization to traffic engineering. The sample data is processed by normalization, dimensionality reduction, discretization, and symbolization based on the three steps of symbolic aggregate approximation (SAX) with driving data characteristics taken into adequate consideration. The results showed that the high-dimensionality miscellaneous driving time series data was rationally converted into highly readable, easy to search and locate symbolic series after semantic encoding, and the main characteristics of time series data was preserved after a substantial reduction of data dimensionality. Finally, the paper demonstrated the positive effects of this method on the analysis of actual vehicle driving safety based on case study, and it explored the application of SAX to speed and acceleration data from driving data set.

Inappropriate speed in negotiating curves is the primary cause of rollovers and sideslips. In this study, the authors proposed an improved curve speed model considering driving styles, as well as vehicle and road factors. On the basis of a vehicle-road interaction model, the driver behaviour factor was introduced to quantify driving styles of curve speed choices. Firstly, the fuzzy synthetic evaluation method was utilised to classify the driving styles of 30 professional drivers into three different types (i.e. cautious, moderate and aggressive). Secondly, the classification results using fuzzy synthetic evaluation were compared to and verified with the K-means clustering method resulting over 60% the similarities. Finally, the proposed curve speed model was built and compared with four existing models. The authors' model has the following promising advantages: (i) it reflects the speed preferences of three different types of drivers on the premise of driving safety on curves; and (ii) it shows a stationary speed transition when the road adhesion coefficient exceeds 0.8, which indicates that rollover, instead of sideslip, becomes the primary cause for lateral instability crashes on curves. Therefore, this proposed curve speed model could be applied in a curve speed warning system to improve both driving safety and comfort.

This study presents a numerical model that describes the dynamic process of building situation awareness after an automation-initiated transition. The model predicts the level of situation awareness as a function of elapsed time since the transition, and is verified using data from an experiment in which participants watched animated video clips of automated driving scenarios. Additionally, the ‘number of fixations per second’ is suggested for real-time monitoring of situation awareness in automated driving.

Drivers of automated cars may occasionally need to take back manual control after a period of inattentiveness. At present, it is unknown how long it takes to build up situation awareness of a traffic situation. In this study, 34 participants were presented with animated video clips of traffic situations on a three-lane road, from an egocentric viewpoint on a monitor equipped with eye tracker. Each participant viewed 24 videos of different durations (1, 3, 7, 9, 12, or 20 s). After each video, participants reproduced the end of the video by placing cars in a top-down view, and indicated the relative speeds of the placed cars with respect to the ego-vehicle. Results showed that the longer the video length, the lower the absolute error of the number of placed cars, the lower the total distance error between the placed cars and actual cars, and the lower the geometric difference between the placed cars and the actual cars. These effects appeared to be saturated at video lengths of 7–12 s. The total speed error between placed and actual cars also reduced with video length, but showed no saturation up to 20 s. Glance frequencies to the mirrors decreased with observation time, which is consistent with the notion that participants first estimated the spatial pattern of cars after which they directed their attention to individual cars. In conclusion, observers are able to reproduce the layout of a situation quickly, but the assessment of relative speeds takes 20 s or more.