No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Predicting Take-Over Times of Truck Drivers in Conditional Autonomous Driving
Abstract
Conditional autonomous driving requires the description of sufficient time reserves for drivers in take-over situations. The definition of this time reserve has not been addressed for the truck context thus far. Through the observation of physiological measures, the possibility of estimating reaction times is considered. Driver data is collected with a remote eye-tracker and body posture camera. Empirical data from a simulator study is utilized to train and compare four machine learning algorithms and generate driver features. The estimation of take-over times is defined as a classification problem with four reaction time classes, leading to a misclassification rate of a linear support vector machine (SVM) of 38.7%. Utility of driver features for reaction time estimation are discussed.
... Since understanding of the influencing factors on the driver's reaction is needed, better measurement 65 techniques and algorithms are required that potentially predict the right time interval for secure human 66 behavior (Lotz, et al., 2019). The second approach for the definition of adequate take-over times is a more 67 elegant, however, technically more complex due to additional sensors and algorithms necessary for driver 68 observation. ...
... This incorporates the data collection of reaction times and 258 eye-tracking data. The measurements of the body posture are not included in this publication but are 259 considered in Lotz et al. (2019). A moving-based simulator will be used to manipulate environmental factors, 260 ...
The desire to enable conditional automated driving (CAD) in the near future, entails the challenge to manage drivers’ safe transitions from automation back to manual control. Several factors have been considered in recent years in the passenger car context, while the truck has largely been disregarded. For the first time take-over behavior of heavy-duty truck drivers in time critical take-overs is considered in CAD research. This study analyzes the effect of non-driving related tasks, CAD duration, take-over situations and number of take-overs on reaction times of truck drivers. Gaze behavior was tracked with a remote eye-tracker; reaction times and driver interaction during CAD drives was recorded and analyzed. Two different non-driving related tasks were presented in nine unique take-over situations, while also controlling for the duration of CAD. Contrary to assumption, no influence of non-driving related tasks or CAD duration on reaction times is found. Notably, different reaction times are recorded due to the nine unique take-over situations. Finally, it is shown that our take-over times decrease over the course of the experiment and are far lower than other published reaction times (M = 1.35 s) in the passenger car context. The findings are discussed and implications with regard to other published studies are drawn.
... The driver's reaction time in three typical collision scenarios with controlled TTC illustrated that the drivers' steering manoeuvre reaction was significantly faster than the brake reaction time with the mean of 0.86 s with comparison to 2.29 s for braking reaction (Li et al., 2019). Another study in the driving simulator for the takeover action (TOR) of truck drivers in autonomous driving (level 3) by using eye tracker from 755 TOR found braking pedal reaction time lower than 1.7 s (Lotz et al., 2019). The initial reaction braking delay of 2.8 s is being considered in the calculation of the braking distance in the Italian road design code, which decreases 1% with the speed of the driver. ...
The related research activities were carried out in collaboration with the University of Bologna, Paris-Est University and Gustave Eiffel University (IFSTTAR) in the form of a cotutelle PhD. The activities are divided into two macro areas ; the driving simulation studies conducted in Gustave Eifel University (IFSTTAR) and on-road experiments organized by the University of Bologna. The first part of the research is focused on improving the physical fidelity of the two DOF driving simulator with particular attention to motion cueing and vehicle dynamics model. The vehicle dynamics model has been developed in MATLAB-Simulink and has the ability of real-time calculation of the vehicle states and control the motion platform. During this phase of the research, motion cueing algorithms were developed to control the simulator movements and the effect of the motion cues on drivers’ behaviour was analysed through experimentation. The results of these studies are discussed in the case studies I and II. In the second part of the research, the driver performance and visual behaviour were studied on the road under different scenarios. The driver visual behaviour was recorded with the use of a head mounted eye-tracking device, while the vehicle trajectory was registered with an instrumented vehicle equipped with Global Positioning System (GPS). During this phase, several case studies were developed to monitor drivers’ behaviour in the naturalistic environment. Case study III aims to integrate the traditional road safety auditing with an innovative driver behaviour monitoring system. The real road experiment with drivers was carried out in an urban arterial road in order to evaluate the proposed approach through innovative driver monitoring techniques. These same driving monitoring instruments were used for evaluating the improvement of a pedestrian crossing at the roundabout in case study IV. The eye-tracking data were evaluated in both studies in order to identify a driver visual attention indicator based on the participants gaze position and duration. Significant attention is given to the safety of vulnerable drivers in urban areas during the naturalistic driving behaviour study. Case study V analyzed the driver yielding behaviour in approach phase to a bicycle priority crossing with the use of surrogate safety measures. The drivers’ performance measures such as perception reaction time and gaze behaviour were used to assess the safety level of the crossing equipped with standard and innovative signalling systems. The improvement on the driver’s yielding behaviour towards an un-signalized crossing during night-time and their reaction to an integrated lighting-warning system was evaluated in the case study VI. The last phase of the thesis is dedicated to the study of Adaptive Cruise Control (ACC) with on-road and simulator experimentation. The on-road experimentation investigated the driver assistant system influence on the drivers' adaptation with objective and subjective assessment, in which an eye tracking instrument and EEG helmet were used to monitor the drivers on a highway. The results are presented in Case studies VII and VIII and drivers’s visual attention was reduced due to adaptation to the ACC in the car following scenario. The results of the on-road test were later used to reproduce to the same scenario in the driving simulator and the adaptation of drivers’ behaviour with the use of ACC was confirmed through experimentation
... In terms of driver take-over prediction, Lotz et al. was able to predict a drivers take-over time within four discrete ranges, in seconds [9]. Gold et al. improved upon Lotz's work by modeling driver take-over time as a continuous variable through the use of a generalized non-linear regression model [10]. ...
The goal of this study is to provide a framework, using hidden semi-Markov models, for modeling a driver's response time after an alert is provided in manual driving. Given the plethora of alerts and warning within a vehicle, there is a need to understand when a driver will respond after an alert is provided. Data from a previous driving simulator study, where drivers were interacting with an in-vehicle information system (IVIS) were used for model training. The final data set included 16 participants, with 288 task initiations. The proposed model could predict a driver's response time accurately using only a small portion of the available data, and had a mean absolute error of 0.51 seconds with 84% of predictions within an absolute error of 1 second. This framework has applicability in mitigating the risk of transitions in driver distraction. This includes transitions from the road to the secondary task and back to the road.
... regression models, to extrapolate data based on empirical findings post-hoc and explain correlations in the data (McDonald, et al., 2019;Zhang, et al., 2019). A second class of models provides online prediction based on data obtained through driver and environment monitoring (Nilsson et al., 2015;Braunagel et al., 2017;Lotz & Weissenberger, 2019). However, subjective driver interpretation is missing as input data. ...
Partial and conditional automated driving allows the driver to transfer responsibility to the vehicle. While assistance systems are designed to deal with aspects of the driving task, currently no assistance systems are available to predict driver behaviour for take-over when the vehicle is handling the driving task. This is important asdrivers might interpret driving situations differently than an activated automation function. This can cause self-initiated take-overs leading to a reduction of trust in the system.In theory, if a prediction is robust, an assistance system could also adapt basedon this prediction.Anew subjective complexity model addressing these situationsis introduced. The subjective complexity model learns situations in which individualdrivershavepreviously self-initiated control of the driving task. Based onexemplary sideswipe manoeuvres,the system concept is explained and simulated with a training and test dataset.Upon introducing this system,a discussion is initated on the difference between objective and subjective situationcomplexity. A distinction is drawn between mathematical descriptions based on vehicular sensor data and human interpretation of the environment. The proposed system also functions as a carrier technology for further investigations between the differences of objective and subjective complexities.
With the introduction of advanced driving assistance systems managing longitudinal and lateral control, conditional automated driving is seemingly in near future of series vehicles. While take-over behavior in the passenger car context has been investigated intensively in recent years, publications on semi-trucks with professional drivers are sparse. The effects influencing expert drivers during take-overs in this context lack thorough investigation and are required to design systems that facilitate safe take-overs. While multiple findings seem to cohere in passenger cars and semi-trucks, these findings rely on simulated studies without taking environments as found in the real world into account. A test track study was conducted, simulating highway driving with 27 professional non-affiliated truck drivers. The participants drove an automated Level 3 semi-truck while a non-driving-related task was available. Multiple time critical take-over situations were initiated during the drives to investigate four main objectives regarding driver behavior. (1) With these results, comparison of reaction times and behavior can be drawn to previous simulator studies. The effect of situation criticality (2) and training (3) of take-over situations is investigated. (4) The influence of warning expectation on driver behavior is explored. Results obtained displayed very quick time to hands on steering and time to first reaction all under 2.4 s. Highly critical situations generate very quick reaction times M = 0.81 s, while the manipulation of expectancy yielded no significant variation in reaction times. These reaction times serve as a reference of what can be expected from drivers under optimal take-over conditions, with quick reactions at high speed in critical situations.
The driver of a conditionally automated car is not required to permanently monitor the outside environment, but needs to take over control whenever the automation issues a “request to intervene” (i.e., take-over request). If the driver misses the take-over request or does not respond in a timely and correct manner, a takeover could result in a safety-critical scenario. Traditionally, warnings in vehicles are conveyed by visual and auditory displays, though recently it has been argued that vibrotactile stimuli could also be a viable approach to present a take-over request to the driver.
In this paper, we present a vibrotactile seat designed to convey dynamic vibration patterns to the driver. The seat incorporates 48 vibration motors (eccentric mass rotation) that can be individually controlled. One 6x4 matrix, with an average in-ter-motor distance of approximately 4 cm, is located in the seat back and one in the seat bottom. The DC-voltage to the motors is controlled by three Pulse Width Modulation (PWM) drivers, which in turn are controlled by an Arduino micro-controller.
A study with 12 participants was conducted to investigate (1) at which vibration intensity participants find a vibratory stimulus annoying and whether this thresh-old changes over time, (2) how well participants are able to discriminate vibratory stimuli as a function of spatial separation, and (3) which of six dynamic vibration patterns are regarded as most satisfying.
Results showed that participants’ annoyance threshold reduced when they were repeatedly exposed to vibrotactile stimuli. Second, the percentage of correct re-sponses in the two-point discrimination test increased significantly with increas-ing inter-stimuli distance (i.e., from 4 to 20 cm). Third, participants seemed to be more satisfied when more motors were activated simultaneously (i.e., more spatial overlap).
Overall, the results suggest that participants are well able to perceive vibrotactile in the driver seat. However, the results suggest that repetitive exposure to vi-brotactile stimuli may evoke annoyance, a finding that should be taken into ac-count in future designs of vibrotactile displays. Future studies should investigate the possibility to convey complex messages via the vibration seat.
Highly automated driving constitutes a temporary transfer of the primary driving task from the driver to the automated vehicle. In case of system limits, drivers take back control of the vehicle. This study investigates the effect of varying traffic situations and non-driving related tasks on the take-over process and quality. The experiment is conducted in a high-fidelity driving simulator. The standardized visual Surrogate Reference Task (SuRT) and the cognitive n-back Task are used to simulate the non-driving related tasks. Participants experience four different traffic situations. Results of this experiment show a strong influence of the traffic situations on the take-over quality in a highway setting, if the traffic density is high. The non-driving related tasks SuRT and the n-back Task show similar effects on the take-over process with a higher total number of collisions by the SuRT in the high density traffic situation.
In recent years, one of the goals of driver assistance systems has been to relieve the driver of parts of his burden of driving. Current developments are able to take over full control of the vehicle in specific use cases. Automated driving in traffic jams is seen as the next step towards automated driving. One of the main challenges is ensuring the driver's ability to take back responsibility when being driven by the car. The assumption that drivers tend to perform secondary tasks creates a target conflict. Within this context, it is the aim of the current study to analyze the impacts of an exemplary secondary task and certain Human-Machine Interaction (HMI) mechanisms on the driver's level of situation awareness and the resulting driving safety. The implemented HMI mechanisms are a “driver's safety device” and a “video image”. While it is hypothesized that the secondary task has a negative influence on situation awareness, the HMI mechanism's effect is supposed to be positive. In addition, the overall user's acceptance will also be analyzed. The results show that the implemented driver's safety device has no effect on situation awareness but is still able to support driving safety by assisting the driver in the execution of actions. The video image has a positive effect on situation awareness but because the complacency-effect is enhanced its effect on driving safety is negligible. As assumed, the secondary task has a negative impact on situation awareness and driving safety.
We review accuracy estimation methods and compare the two most common methods: crossvalidation and bootstrap. Recent experimental results on arti cial data and theoretical results in restricted settings have shown that for selecting a good classi er from a set of classiers (model selection), ten-fold cross-validation may be better than the more expensive leaveone-out cross-validation. We report on a largescale experiment|over half a million runs of C4.5 and a Naive-Bayes algorithm|to estimate the e ects of di erent parameters on these algorithms on real-world datasets. For crossvalidation, we vary the number of folds and whether the folds are strati ed or not � for bootstrap, we vary the number of bootstrap samples. Our results indicate that for real-word datasets similar to ours, the best method to use for model selection is ten-fold strati ed cross validation, even if computation power allows using more folds. 1
This paper presents a theoretical model of situation awareness based on its role in dynamic human decision making in a variety of domains. Situation awareness is presented as a predominant concern in system operation, based on a descriptive view of decision making. The relationship between situation awareness and numerous individual and environmental factors is explored. Among these factors, attention and working memory are presented as critical factors limiting operators from acquiring and interpreting information from the environment to form situation awareness, and mental models and goal-directed behavior are hypothesized as important mechanisms for overcoming these limits. The impact of design features, workload, stress, system complexity, and automation on operator situation awareness is addressed, and a taxonomy of errors in situation awareness is introduced, based on the model presented. The model is used to generate design implications for enhancing operator situation awareness and future directions for situation awareness research.
We review accuracy estimation methods and compare the two most common methods: crossvalidation and bootstrap. Recent experimental results on artificial data and theoretical results in restricted settings have shown that for selecting a good classifier from a set of classifiers (model selection), ten-fold cross-validation may be better than the more expensiveleaveone -out cross-validation. We report on a largescale experiment---over half a million runs of C4.5 and a Naive-Bayes algorithm---to estimate the effects of different parameters on these algorithms on real-world datasets. For crossvalidation, wevary the number of folds and whether the folds are stratified or not# for bootstrap, wevary the number of bootstrap samples. Our results indicate that for real-word datasets similar to ours, the best method to use for model selection is ten-fold stratified cross validation, even if computation power allows using more folds. 1 Introduction It can not be emphasized eno...
The next step towards the fully automated vehicle is the level of conditional automation, where the automated driving system can take over the control and responsibility for a limited time interval. Nevertheless, take-over situations may occur, forcing the driver to resume the driving task. Despite such situations, the driver is able to perform secondary tasks during conditionally automated driving, hence a low take-over quality must be expected. Methods for Driver-Activity Recognition (DAR) usually extract features for the classification within a moving time window. In this paper, the first DAR architecture based on the driver's scanpath, which is extracted by means of dynamic clustering and symbolic aggregate approximation patterns, is introduced. To demonstrate the potential of this approach, it is compared to a state-of-the-art method based on the data of a driving simulator study with 82 subjects. The classification performance of both DAR approaches was examined for decreasing window sizes with regard to the recognition of different secondary tasks and the separability of drivers using a handheld or hands-free device. Compared to the state-of-the-art approach, the proposed method shows a classification accuracy increase of nearly 20%, a significant improvement of the overall classification performance, and is able to classify the secondary tasks of the driver even for short windows of a duration of 5 s, i.e. with little information.
Recent studies analyzing driver behavior report that various factors may influence a driver's take-over readiness when resuming control after an automated driving section. However, there has been little effort made to transfer and integrate these findings into an automated system which classifies the driver's take-over readiness and derives the expected take-over quality. This study now introduces a new advanced driver assistance system to classify the driver's takeover readiness in conditionally automated driving scenarios. The proposed system works preemptively, i.e., the driver is warned in advance if a low take-over readiness is to be expected. The classification of the take-over readiness is based on three information sources: (i) the complexity of the traffic situation, (ii) the current secondary task of the driver, and (iii) the gazes at the road. An evaluation based on a driving simulator study with 81 subjects showed that the proposed system can detect the take-over readiness with an accuracy of 79%. Moreover, the impact of the character of the take-over intervention on the classification result is investigated. Finally, a proof of concept of the novel driver assistance system is provided showing that more than half of the drivers with a low take-over readiness would be warned preemptively with only a 13% false alarm rate.
This paper discusses the ways in which automation of industrial processes may expand rather than eliminate problems with the human operator. Some comments will be made on methods of alleviating these problems within the 'classic' approach of leaving the operator with responsibility for abnormal conditions, and on the potential for continued use of the human operator for on-line decision-making within human-computer collaboration.
When reaching a system limit, highly automated vehicles arbitrate control back to the driver. In some situations, a complete arbitration to the driver may not be appropriate or favored and a monitoring of the system sufficient. This study verifies the opportunity to use partial automation as a fallback level of high automation. Furthermore, monitoring with hands on the steering wheel is compared to monitoring solely visually. A study with 32 subjects and another 16 subjects in the baseline condition was conducted in the dynamic driving simulator of the BMW Research and Technology. In two out of six situations, subjects had to take over control after monitoring the system for two seconds in a time critical situation. A faster intervention with less braking and acceleration was observed in the baseline group compared to the automated condition. Nevertheless, concerning the subjective rating of the subjects, the partial automation was considered as comfortable and useful. Concerning the type of monitoring, hands on the steering wheel led to a 0.3 s faster intervention without statistical significance (p=0.158; r=0.234).
The goal of this paper is to advance the state-of-the-art of articulated pose estimation in scenes with multiple people. To that end we contribute on three fronts. We propose (1) improved body part detectors that generate effective bottom-up proposals for body parts; (2) novel image-conditioned pairwise terms that allow to assemble the proposals into a variable number of consistent body part configurations; and (3) an incremental optimization strategy that explores the search space more efficiently thus leading both to better performance and significant speed-up factors. Evaluation is done on two single-person and two multi-person pose estimation benchmarks. The proposed approach significantly outperforms best known multi-person pose estimation results while demonstrating competitive performance on the task of single person pose estimation (Models and code available at http:// pose. mpi-inf. mpg. de).
The goal of this paper is to advance the state-of-the-art of articulated pose estimation in scenes with multiple people. To that end we contribute on three fronts. We propose (1) improved body part detectors that generate effective bottom-up proposals for body parts; (2) novel image-conditioned pairwise terms that allow to assemble the proposals into a variable number of consistent body part configurations; and (3) an incremental optimization strategy that explores the search space more efficiently thus leading both to better performance and significant speed-up factors. We evaluate our approach on two single-person and two multi-person pose estimation benchmarks. The proposed approach significantly outperforms best known multi-person pose estimation results while demonstrating competitive performance on the task of single person pose estimation. Models and code available at http://pose.mpi-inf.mpg.de
In this paper, a multiview, multimodal vision framework is proposed in order to characterize driver activity based on head, eye, and hand cues. Leveraging the three types of cues allows for a richer description of the driver's state and for improved activity detection performance. First, regions of interest are extracted from two videos, one observing the driver's hands and one the driver's head. Next, hand location hypotheses are generated and integrated with a head pose and facial landmark module in order to classify driver activity into three states: wheel region interaction with two hands on the wheel, gear region activity, or instrument cluster region activity. The method is evaluated on a video dataset captured in on-road settings.
This paper discusses the ways in which automation of industrial processes may expand rather than eliminate problems with the human operator. Some comments will be made on methods of alleviating these problems within the 'classic' approach of leaving the operator with responsibility for abnormal conditions, and on the potential for continued use of the human operator for on-line decision-making within human-computer collaboration.
This test-track study assessed whether adaptive cruise control (ACC) induces behavioural adaptation in drivers. Eighteen experienced drivers drove a test vehicle while following a lead vehicle in three counterbalanced conditions: No ACC (self-maintained average headway of 2 s), ACC-Short (headway of 1.4 s) and ACC-Long (headway of 2.4 s). Results demonstrate that ACC can induce behavioural adaptation in drivers in potentially safety-critical ways. Compared to driving unsupported, participants located significantly more items per minute on a secondary task when using ACC, while their response times to a hazard detection task increased. This effect was particularly pronounced in those scoring high on a sensation-seeking scale. Using ACC resulted in significantly more lane position variability, an effect that was also more pronounced in high sensation-seekers. Drivers' trust in ACC increased significantly after using the system, and these ratings did not change despite a simulated failure of the ACC system during the ACC-Long condition. Response time to the simulated ACC failure was related to a driver's locus of control: Externals intervened more slowly than Internals. All drivers reported relying on the ACC system to keep their vehicle at a safe distance from the lead vehicle. Results are consistent with similar research conducted on lane departure warning systems. Driver awareness training is a potential preventive strategy that could minimize the behavioural adaptation associated with novel in-vehicle systems such as ACC.
Reports of 3 experiments testing the hypothesis that the average duration of responses is directly proportional to the minimum average amount of information per response. The results show that the rate of performance is approximately constant over a wide range of movement amplitude and tolerance limits. This supports the thesis that "the performance capacity of the human motor system plus its associated visual and proprioceptive feedback mechanisms, when measured in information units, is relatively constant over a considerable range of task conditions." 25 references. (PsycINFO Database Record (c) 2006 APA, all rights reserved).
Verkehr auf einen Blick
- Statistisches Bundesamt
Statistisches Bundesamt. Verkehr auf einen Blick, Wiesbaden, Germany (2013)
Uebergabe von hochautomatisiertem Fahren zu manueller Steuerung
- T Vogelpohl
- M Vollrath
- M Kuehn
Vogelpohl, T., Vollrath, M., Kuehn, M.: Uebergabe von hochautomatisiertem Fahren zu
manueller Steuerung. Forschungsbericht 47, (2017)
Automationseffekte im Fahrzeug - von der Reaktion zur Übernahme
- D Damböck
Damböck, D.: Automationseffekte im Fahrzeug-von der Reaktion zur Übernahme. München, Technische Universität München, Dissertation (2013)
ProceedThis is a pre-print of an article published in Advances in Intelligent Systems and Computing. The final authenticated version is available online at
- J Radlmayr
- C Gold
- L Lorenz
- M Farid
- K Bengler
Radlmayr, J., Gold, C., Lorenz, L., Farid, M., Bengler, K.: How traffic situations and nondriving related tasks affect the take-over quality in highly automated driving. In: ProceedThis is a pre-print of an article published in Advances in Intelligent Systems and Computing.
The final authenticated version is available online at: https://doi.org/10.1007/978-3-319-93885-1_30
ings of the Human Factors and Ergonomics Society 58th Annual Meeting, pp. 2063-2067
(2014)
Automationseffekte im Fahrzeug -von der Reaktion zur Übernahme. München
- D Damböck
Damböck, D.: Automationseffekte im Fahrzeug -von der Reaktion zur Übernahme. München, Technische Universität München, Dissertation (2013)