A multidisciplinary team of TU Delft mobility researchers, known as The Mobilisers, embarked on a journey to explore the future of mobility and how it is embedded in various socio-technological contexts. We developed four radical, yet realistic scenarios for the Dutch Mobility System in 2050, taking into account the Netherlands' position as part of Europe and its role as an international hub. @en

Whereas driver fitness is widely recognized as a prerequisite for safety, the construct lacks a formal framework. We present the first steps towards its operationalization. We interpret availability and allocation of cognitive and physiological resources as fitness dimensions, and risk factors fatigue/drowsiness, distraction/inattention, intoxication, sudden incapacitation and speeding as state variables loading on these dimensions. We collect and synthesize US crash data, and calculate Relative Risks RR and Population Attributable Fractions AFp. Sudden incapacitation (RR = 32.112) is the most detrimental to individual safety, followed by alcohol intoxication (RR=11.277), fatigue/drowsiness (RR=4.966), speeding (RR=2.743), and inattention/distraction (RR=0.241). Taking into account prevalence, alcohol intoxication has the largest impact (AFp=0.068), followed by speeding (AFp=0.065), fatigue/drowsiness (AFp=0.059), incapacitation (AFp=0.013) and inattention/distraction (AFp=−0.416). Alcohol intoxication plays a major role in fatal crashes (RR=47.341, AFp=0.247), followed by speeding (RR=9.364, AFp=0.25). The analysis emphasizes the dangers of intoxicated driving and speeding, but also reveals shortcomings in census data, notably an under-representation of inattention, as well as the need for specific data collection on intoxication, reckless driving and sudden incapacitation in crashes. Taken together, the conceptual model and data synthesis provide a first step towards a framework of driver fitness; formalizing hypotheses on causal relations and providing crude estimates of factor loadings. This framework has practical applications as a model for multimodal driver monitoring systems, and to calculate risk factor impacts to inform policy makers.@en

With increasing implementation of automated driving technology it is expected that different automation modes will be present within the same vehicle and within a single trip. At all times during automated driving the driver needs to have ‘mode awareness’, which is an understanding of the automation mode and the corresponding responsibilities. Yet, research on HMI design to support mode awareness for multiple automation modes within a single vehicle and within a single trip is currently limited. The current work describes the development and evaluation of a Human Machine Interface (HMI) to support mode awareness while driving in different automation modes. The work exists of three phases: Phase 1 defines functional requirements for HMI design based on literature review and 5 experimental studies including 146 participants. Phase 2 implements the functional requirements in HMI design through expert and focus group sessions. Phase 3 evaluates and improves upon the HMI design employing virtual reality and the RITE (Rapid Iterative Testing and Evaluation) method with 18 participants. The result is a continuous and holistic HMI design creating mode awareness through ambience. Findings from Phase 3 and previous research indicate that this HMI is comprehended well, with a relatively low task load, and with a good experienced system usability. It is important to additionally evaluate the HMI design resulting from the current study in driving simulators and in on-road tests. Such tests will provide an opportunity to verify and expand on the current study's findings and to contribute to guidelines for HMI design.

@en

Higher levels of vehicle automation come with new challenges for designing safe systems. The Human Machine-Interface (HMI) plays a key role in mediating the interaction between the human driver and vehicle automation. By providing the driver with appropriate feedback, the HMI has the potential to increase mode awareness and situational awareness. For the development of appropriate HMI solutions, usability assessments are essential. Immersive Virtual Reality (VR) technology enables researchers and designers to construct realistic virtual prototypes and immersive evaluation scenarios with less time and resources. The current study presents a VR evaluation tool called VRHEAD, which is designed to facilitate an iterative design process and support the rapid implementation of virtual prototypes to evaluate of an automated vehicle's HMI. Initial results indicate that VRHEAD is a promising approach for the rapid implementation and evaluation of design concepts. The use of VR tools, like VRHEAD, can reduce the time and costs associated with developing high-fidelity prototypes and provide more flexibility in modifying a design according to new research findings, thus broadening the exploration of the HMI design space.

@en

This study reports usage of supervised automation and driver attention from longitudinal naturalistic driving observations. Automation inexperienced drivers were provided with instrumented vehicles with adaptive cruise control (ACC) and lane keeping (LK) features (SAE level 2). Data was collected comparing one month of driving without support to two months where drivers were instructed to use automation as desired. On highways, level 2 automation was used respectively 63% and 57% of the time by Tesla and BMW users, with peak usage during slow stop-and-go traffic (0–30 km/h) and higher speeds (>80 km/h). On roads with speed limits below 70 km/h, automation was used less than 8%, and use on urban roads was incidental rather than habitual. Automation usage increased with time in trip, but no clear time of day effects were found. Head pose data could not classify driver attention, and we recommend gaze tracking in future studies. Head pose deviation was selected as alternative indicator for monitoring activity. Comparing among forms of automation usage on the highway, head heading deviation was smallest during ACC use, but did not differ between automation and baseline manual driving. Head heading deviation during manual driving was smaller in the baseline than the experimental phase, which suggests that motives for manual highway driving may be attention related. Automation usage did not change much over the first 12 weeks of the experimental condition, and there were no longitudinal changes in head pose deviation.

@en

In the transition towards higher levels of vehicle automation, one of the key concerns with regards to human factors is to avoid mode confusion, when drivers misinterpret the driving mode and therewith misjudge their own tasks and responsibility. To enhance mode awareness, a clear human centered Human Machine Interface (HMI) is essential. The HMI should support the driver tasks of both supervising the driving environment when needed and self-regulating their non-driving related activities (NDRAs). Such support may be provided by either presenting continuous information on automation reliability, from which the driver needs to infer what task is required, or by presenting continuous information on the currently required driving task and allowed NDRA directly. Additionally, it can be valuable to provide continuous information to support anticipation of upcoming changes in the automation mode and its associated reliability or required and allowed driver task(s). Information that could support anticipation includes the available time until a change in mode (i.e. time budget), information on the upcoming mode, and reasons for changing to the upcoming mode. The current work investigates the effects of communicating this potentially valuable information through HMI design. Participants received information from an HMI during simulated drives in a simulated car presented online (using Microsoft Teams) with an experimenter virtually accompanying and guiding each session. The HMI either communicated on automation reliability or on the driver task, and either included information supporting anticipation or did not include such information. Participants were thinking aloud during the simulated drives and reported on their experience and preferences afterwards. Anticipatory information supported understanding about upcoming changes without causing information overload or overreliance. Moreover, anticipatory information and information on automation reliability, and especially a combination of the two, best supported understandability and usability. Recommendations are provided for future work on facilitating supervision and NDRA self-regulation during automated driving through HMI design.@en

Advanced driver assistance systems such as adaptive cruise control (ACC) and lane keeping system (LKS) potentially contribute to reducing crash rates and traffic congestion. On-road studies based on early ACC systems operational at medium–high speeds only have shown that the system reduces the proportion of short time gaps when activated. Despite the effects on driver behaviour, most mathematical models assessing the impact of ACC and LKS systems on crash rates and traffic congestion are not based on empirical findings. This study examines the factors that influence changes in the longitudinal vehicle control when driving with ACC and LKS. The data were collected in a naturalistic driving experiment with full-range ACC and LKS and two different vehicle brands (BMW and Tesla) in the Netherlands. To capture changes that are relevant for traffic safety, speeding and a time gap shorter than one second were investigated. The factors influencing speeding and short time gaps were analysed using statistical tests and logistic regression models with random effects, that allow to control for the impact of different explanatory variables and correlations between repeated 10-s intervals over time. The findings revealed that, overall, drivers were less likely to speed and they were also less likely to have a time gap shorter than one second in the experimental condition with the ACC and the LKS than in the baseline condition in manual driving. Drivers were likely to speed in the following 10-s interval when the current speed was close to the speed limit, and/or when the next speed limit was lower than the current speed limit, and/or when the acceleration was high. Drivers were likely to have a short time gap in the following 10-s interval when approaching a slower leader, and/or when the current time gap was short and/or when the acceleration was high. Controlled for these main factors, drivers were less likely to speed and to have a short time gap when the ACC and the LKS were active. However, drivers were more likely to speed when overruling the ACC by pressing the gas pedal. When the systems were active, one vehicle brand showed a smaller probability of a short time gap than the other brand, suggesting differences in ACC system settings between brands. In addition, the speeding probability increased while the probability of a short time gap decreased over time during the trip after the activation of the systems. Although further studies including a larger sample of participants and a wider range of traffic situations are needed, the results are useful to the design of automated vehicles that prevent speeding and short time gaps, and to the implementation of traffic simulations that evaluate the impact of ACC and LKS on crash rates and traffic congestion according to realistic on-road data.

@en

Automatic incident detection (AID) systems and variable speed limits (VSLs) can reduce crash probability and traffic congestion. Studies based on loop detector data have shown that AID systems decrease the variation in speeds between drivers. Despite the impact on driver behaviour characteristics, most mathematical models evaluating the effect of AID systems on traffic operations do not capture driver response realistically. This study examines the main factors related to driver speed compliance with a sequence of three VSLs triggered by an AID system. For this purpose, the variable speed limit database of the executive agency of the Dutch Ministry of Infrastructure and Water Management (Rijkswaterstaat) was integrated into the UDRIVE naturalistic driving database for passenger car data collected in the Netherlands. The video data were annotated to analyse driver glance behaviour and secondary task engagement. A logistic regression model was estimated to predict driver speed compliance after each VSL in the sequence. The results reveal that the factors predicting compliance to the VSLs differ based on which of the three VSLs the driver is subjected to. Low speeds and accelerations before the gantry, approaching a slower leader, high proportion of time with eyes-on-road and close consecutive gantries were associated with high compliance with the first VSL in the sequence (i.e., indicating a speed limit of 70 km/h with flashing attention lights). Low speeds and accelerations before the gantry, close consecutive gantries and a small number of lanes resulted in high compliance with the second VSL (i.e., a speed limit of 50 km/h with flashing attention lights). Low speeds before the gantry and close consecutive gantries were linked to high compliance with the third VSL (i.e., indicating a speed limit of 50 km/h). Although further investigations based on a larger sample are needed, these findings are relevant to the development of human-like driving assistance systems and of traffic simulations that assess the impact of AID systems on traffic operations realistically.

@en

Mental workload plays a central role in driver behavior. Unlike physical workload, mental workload is difficult to quantify as it is the result of the interaction between the task to perform, that is, the task demands, and the capacity to perform, that is, the mental resources. While the first can be quantified, the second—as of today—cannot. Still, within individuals and for homogenous groups, judgments about mental workload can be made. Driver state has a large influence on mental workload, and the two concepts cannot be considered separately: deteriorated driver state increases mental workload as the driver has to invest more effort to maintain performance. This chapter addresses the concepts of driver mental workload and driver state and how to assess them, with special attention to the effects of automation on driver state and workload@en

Traffic congestion and crash rates can be reduced by introducing variable speed limits (VSLs) and automatic incident detection (AID) systems. Previous findings based on loop detector measurements have revealed that drivers reduce their speeds while approaching traffic congestion when the AID system is active. Notwithstanding these behavioural effects, most microscopic traffic flow models assessing the impact of VSLs do not describe driver response accurately. This study analyses the main factors that influence driver deceleration behaviour while approaching traffic congestion with and without VSLs. The Dutch VSL database was linked to the driver behaviour data collected in the UDRIVE naturalistic driving study. Driver engagement in secondary tasks and glance behaviour were extracted from the video data. Linear mixed-effects models predicting the characteristics of deceleration events were estimated. The results show that the maximum deceleration is high when approaching a slower leader, when driving at high speeds and short distance headways, and close to the beginning of traffic congestion. The minimum time headway is short when driving at high speeds and changing lanes. Certain drivers showed higher decelerations and shorter time headways than others. Controlled for these main factors, smaller maximum decelerations were found when the VSLs were present and visible, and when the gantries were within close proximity. These factors could be incorporated into microscopic traffic simulations to evaluate the impact of AID systems on traffic congestion more realistically. Further research is needed to clarify the link between engagement in secondary tasks, glance behaviour and deceleration behaviour.

@en

Vehicle automation technology is progressing rapidly, and vehicle manufacturers now have the capability to manufacture a wholly self-driving vehicle that can cope with the demands of many traffic situations. However, there is still more research required before the reliability, safety and security of self-driving vehicles can be totally assured. Several Human Factors issues are evident, and much research has been devoted recently to issues such as Takeover Requests and Driver-in-the-Loop (e.g.), In-vehicle HMI for self-diving vehicles, Distraction and Inattention and Fatigue amongst many other concerns. [...]@en

Traffic congestion is a major societal challenge. By advising drivers on the optimal lane to drive, traffic flow can be improved, and congestion reduced. In this paper we describe the development of a lane-specific advice Human Machine Interface (HMI). Persuading drivers to follow an advice that is beneficial to the traffic situation, but may not be immediately beneficial to the drivers themselves, is challenging. In this paper we define persuasive elements to encourage drivers to follow the lane-specific advices. We then describe the interface design process, followed by its evaluation using a driving simulator study. In the simulator study, the effect of two types of persuasion are evaluated: a competitive variant where drivers could earn points and compete with others, and a cooperative variant where real-time information on the number of compliant drivers was available. Participants drove in the simulator on two days. Between days, the treatment groups viewed a Web-portal showing their performance and encouragement from an avatar. Those in the competitive and cooperative groups followed significantly more advices (117 and 111) than those in the control group (89). No significant differences were visible between competitive and cooperative groups. The differences between groups only emerged on the second day.@en

Heart rate data are often collected in human factors studies, including those into vehicle automation. Advances in open hardware platforms and off-the-shelf photoplethysmogram (PPG) sensors allow the non-intrusive collection of heart rate data at very low cost. However, the signal is not trivial to analyse, since the morphology of PPG waveforms differs from electrocardiogram (ECG) waveforms and shows different noise patterns. Few validated open source available algorithms exist that handle PPG data well, as most of these algorithms are specifically designed for ECG data.

In this paper we present the validation of a novel algorithm named HeartPy, useful for the analysis of heart rate data collected in noisy settings, such as when driving a car or when in a simulator. We benchmark the performance on two types of datasets and show that the developed algorithm performs well. Further research steps are discussed.@en

This paper describes the functioning and development of HeartPy: a heart rate analysis toolkit designed for photoplethysmogram (PPG) data. Most openly available algorithms focus on electrocardiogram (ECG) data, which has very different signal properties and morphology, creating a problem with analysis. ECG-based algorithms generally don't function well on PPG data, especially noisy PPG data collected in experimental studies. To counter this, we developed HeartPy to be a noise-resistant algorithm that handles PPG data well. It has been implemented in Python and C. Arduino IDE sketches for popular boards (Arduino, Teensy) are available to enable data collection as well. This provides both pc-based and wearable implementations of the software, which allows rapid reuse by researchers looking for a validated heart rate analysis toolkit for use in human factors studies.

@en

This paper describes the functioning and development of HeartPy: a heart rate analysis toolkit designed for photoplethysmogram (PPG) data. Most openly available algorithms focus on electrocardiogram (ECG) data, which has very different signal properties and morphology, creating a problem with analysis. ECG-based algorithms generally don’t function well on PPG data, especially noisy PPG data collected in experimental studies. To counter this, we developed HeartPy to be a noise-resistant algorithm that handles PPG data well. It has been implemented in Python and C. Arduino IDE sketches for popular boards (Arduino, Teensy) are available to enable data collection as well. This provides both pc-based and wearable implementations of the software, which allows rapid reuse by researchers looking for a validated heart rate analysis toolkit for use in human factors studies.@en

Persuasive in-vehicle systems aim to intuitively influence the attitudes and/or behaviour of a driver (i.e. without forcing them). However, the challenge in using these systems in a driving setting, is to maximise the persuasive effect without infringing upon the driver's safety. This paper proposes a conceptual model for driver persuasion at the tactical level (i.e., driver manoeuvring level, such as lane-changing and car-following). The main focus of the conceptual model is to describe how to safely persuade a driver to change his or her behaviour, and how persuasive systems may affect driver behaviour. First, existing conceptual and theoretical models that describe behaviour are discussed, along with their applicability to the driving task. Next, we investigate the persuasive methods used with a focus on the traffic domain. Based on this we develop a conceptual model that incorporates behavioural theories and persuasive methods, and which describes how effective and safe driver persuasion functions. Finally, we apply the model to a case study of a lane-specific advice system that aims to reduce travel time delay and traffic congestion by advising some drivers to change lanes in order to achieve a better distribution of traffic over the motorway lanes.
@en

Mobile phone use while driving is a major concern for traffic safety. Various studies indicate negative effects of distracted driving and recent Naturalistic Driving studies report substantial increase in crash risk of mobile phone use while driving. The increasing level of vehicle automation is likely to further increase phone use behind the wheel, as the automation takes over part of the driving task drivers are likely to experience boredom and feel more confident to get involved in other activities such as phone use. This may further increase the distraction related accidents on our roads. However, the extent to which this further increases depends largely on the drivers’ level of self-regulation of risk, the timing of engagement in phone activities in relation to the driving context. The objective of this study was to investigate if drivers self-regulate their mobile phone use, specifically focussed on the visual manual (VM) task which is associated with the largest increase in risk, while driving in relation to different driving contexts. For this study naturalistic driving data was analysed of Dutch car drivers collected in the UDRIVE project. The results show that Dutch drivers spent over 9% of all driving time engaging in mobile phone related tasks, including calling and VM tasks such as texting. Drivers used their mobile phone significantly less when a passenger was present. Also, significantly more VM tasks were initiated during standstill than for the other speed categories. In addition, on rural roads relatively less time was spent on VM tasks and on highways relatively more time was spent on VM tasks. Overall the results show that the driving context seems to influence the initiation and involvement in VM tasks, which is an indication that drivers self-regulate this behaviour.

@en

Within the UDRIVE project, a rich cross-European naturalistic driving database was created which includes everyday driving data on car and truck drivers and powered two-wheeler riders. The database provides extensive, reliable insights into driving behavior in real traffic as a foundation for improving the safety and sustainability of European road traffic. This paper discusses the characteristics of the data in the UDRIVE database—elucidating key methodological choices and presenting a selection of results to date. A priority of the study design was obtaining in-depth information on driving behavior, permitting the exploration of diverse research questions. A tailor-made data acquisition system collected very comprehensive data. A total of 287 drivers/riders participated. The sample size restricts the addressable research topics to common behaviors in everyday driving and limits the generalizability of results. However, the data are extensive and promising analyses have already been performed. The results show differences between European countries for distracting activities, seatbelt use, and looking behavior towards cyclists at urban intersections. Moreover, it shows that European drivers engage less in mobile phone use than U.S. drivers. It is likely that European drivers differ in other ways, also—highlighting the dataset's value for developing and implementing targeted safety measures, for the E.U. and its individual countries. Based on the comparison of the different studies, the paper introduces the general conceptual framework for naturalistic driving studies, providing insight in the relation between the scope of a naturalistic driving study and the key methodological choices on sample selection and data acquisition system.

@en

The present study aims to add to the literature on driver workload prediction using machine learning methods. The main aim is to develop workload prediction on a multi-class basis, rather than a binary high/low distinction as often found in litearature. The presented approach relies on measures that can be obtained unobtrusively in the driving environment with off-the-shelf sensors, and on machine learning methods that can be implemented on low-power embedded systems.

Two simulator studies were performed, one inducing workload using realistic driving conditions, and one inducing workload with a relatively demanding lane-keeping task. Individual and group-based machine learning models were trained on both datasets and evaluated. For the group-based models the generalising capability, that is the performance when predicting data from previously unseen individuals, was also assessed.

Results show that multi-class workload prediction on the individual and group level works well, achieving high correct rates and accuracy scores. Generalising between individuals proved difficult using realistic driving conditions, but worked very well in the high demanding lane-keeping task. Reasons for this discrepancy are discussed as well as future research directions.@en