The human-machine interfaces (HMIs) in passenger cars have become more complex over the years, with touch screens replacing physical buttons and more options becoming available. This can result in multiple layers of menus that have to be navigated to perform simple tasks and therefore increase distraction. In addition to this, there is no standardisation for passenger car HMIs. To face these problems, new standards will have to be created. However, knowledge regarding the prevalence of different internal HMI tasks is missing. This complicates efforts to create new standards since it is unclear which internal HMI tasks are most important to assess. It also complicates efforts to create an assessment methodology to rate the safety of different internal HMI interfaces, as tasks that are used more often could be seen as more important. This thesis fills the knowledge gap by identifying the most used internal HMI tasks and the factors that influence their frequencies based on real-world driving observations in a naturalistic setting. In particular, this study especially looks at the influence of car familiarity. Multiple data analysis methods, including Poisson regressions, ANOVA’s, paired t-tests and factor analysis are used to obtain the required knowledge. In general, people seem to perform around 12 tasks per hour in familiar cars versus 9 tasks per hour in unfamiliar cars, but significant differences between tasks exist. The results indicate that the most performed is using the indicator light. Other frequently performed tasks fall into the windshield, radio and media and climate control categories. The factors that influenced the task frequencies included car familiarity, gender, age and weather conditions. The type of car also seems to impact the task frequencies, but more research on this is needed.

This study addresses the critical question: How can AI be effectively harnessed to serve humanity’s global well-being? Referencing Barber’s (1998) dilemma, we ask whether AI will lead us toward a Pandora scenario—marked by chaos and harm—or a Jeffersonian scenario, where AI fosters democratic opportunities. To explore this, we developed a citizen-friendly AI prototype focused on road safety, an urgent global challenge with 1.19 million fatalities annually, particularly in low- and middle-income countries, according to the World Health Organization.

Our AI-co-created website, utilizing Generative AI as an assistant, aims to disseminate proven road safety strategies, such as “Sustainable Safety” and the “Urban Street Design Guide,” which have been successfully implemented in the Netherlands and various U.S. cities. The prototype seeks to bridge language and knowledge gaps, empowering citizens worldwide with accessible road safety insights.

Employing the Prototype Method, this study underwent three development iterations, integrating feedback from diverse citizen groups through workshops and surveys. The final product, prototype 3, presented in a semi-manual, semi-AI format, demonstrates that AI can significantly enhance civic participation and public affairs.

Based on the Human-Centered AI (HCAI) framework, we introduce an advanced "Citizen-AI" model, incorporating the ARIE evaluation model—Avoid, Reduce, Insist, Encourage—to ensure ethical AI deployment. Together with the 3E framework: Education, Empowerment, and Engagement, the ARIE model offers a comprehensive self-assessment tool for developers, guiding the creation of AI systems that prioritize human welfare. This study envisions a future where citizens are equipped to use AI as an active tool for participation, bridging the gap between technology and public engagement.

This dissertation explores what road characteristics trigger drivers’ speed adjustments when approaching freeway curves. It combines speed prediction modelling and human factors research methods. The results show that drivers primarily consider visible cues such as the preceding roadway, deflection angle, and the number of lanes, as opposed to traditional factors like horizontal radius or speed signs, when starting to decelerate. The study advocates for integrating driver perspectives into road design.@en

This dissertation explores cyclist-automated vehicle interactions, emphasising developing and integrating human-machine interfaces (HMIs) to enhance cyclist safety and communication. Adopting a cyclist-centric perspective, it recognises cyclists' unique characteristics and communication strategies in shared traffic environments. Using semi-structured interviews, literature reviews, data triangulation, an eye-tracking field experiment, and a cycling simulator study, the research addresses five key research questions, providing qualitative and quantitative insights.

The main contributions of this dissertation include a thorough investigation of cyclists' expectations for future interactions with automated vehicles, highlighting the need for reliable detection by automated vehicles and placing the responsibility for safety on vehicle developers rather than cyclists. The research offers objective data and self-reported insights into cyclist-automated vehicle interactions and evaluates cyclists' ability to visually detect the presence or absence of a driver. Additionally, it introduces 20 scenarios of cyclist-automated vehicle interaction, serving as a resource for safety assessments and HMI research. A comprehensive literature review of existing HMIs for cyclists was conducted, identifying 92 concepts involving vehicles, bicycles, cyclists, and infrastructure.

The dissertation concludes with design recommendations for cyclist-centric HMIs, proposing an omnidirectional on-vehicle external HMI (eHMI) to communicate detection and automated driving mode. This dissertation provides valuable insights for researchers, policymakers, and automated vehicle developers, aiming for the safer, more inclusive, and sustainable urban traffic environments of tomorrow.@en

This research developed a novel assessment method to enable driving examiners to effectively evaluate the safe use of Advanced Driver Assistance Systems (ADAS) during practical driving exams in the Netherlands. An Assessment Matrix was developed and refined through expert interviews, observations, and case studies. The need to streamline licensing protocols, focus on safety critical competencies and expand examiner training were highlighted in the interviews. ADAS functionality across scenarios and constraints within the driving exam was observed in the case studies. Safety critical competencies identified through observations were monitoring systems, smooth manual takeover, and avoiding distraction. The refined ADAS matrix enabled standardized evaluation despite operational constraints. The findings emphasized integrating ADAS assessments into existing exams, hands-on examiner training, and public education to address knowledge gaps.Recommendations included streamlining assessments, evaluating overall competence, aligning training, providing immersive examiner education, and collaborations to match training with vehicle automation advances. Limitations included sample size and generalizations. For the first time, an empirically validated ADAS evaluation matrix was developed to promote integrating safety critical ADAS competency within the constraints of the practical driving exam. Further research could build on this to refine protocols and ensure that drivers acquired the needed skills as vehicle automation advances.

Automated vehicles with conditional driving automation (SAE level 3 (SAE, 2018)) will request the human driver to intervene when reaching its system boundaries by issuing a take-over request (TOR). This study is investigating whether a speech-based auditory take-over request is influencing the time it takes from automated to manual driving, taking into account the personality trait of the human driver based on theory of Goldberg (1992). The audible warning is based on a woman's voice, varying in three levels of urgency, speech-rate and syntax, and incorporate a lateral deviation measurement by varying the lane width. Due to the COVID-19 pandemic, the experiment was changed to a N=1-study, meaning that only one participant, namely the lead researcher, partook in his own experiment. The driving experiment consisted of 81 runs, each having a TOR after approximately 8 minutes of automated driving. When the automated vehicle is in control, the human driver is asked to do a secondary task, namely the challenging game Tetris on a tablet to get distracted from the situation on the road. It was found that an increase in urgency (take-over type) means a decrease in take-over time (TOT). No significant differences were found for the speech-rate in relation to the TOT, whereas for the syntax, only the STR and UTR had significant differences. Lateral deviation was found to increase when urgency increases, which means that accuracy decreases with higher urgency. Overall, a final design is given based on the results of the N=1-study which could be used for a larger experiment including the personality trait.

The development of automated vehicles on the road is in full swing. As vehicles are getting increasingly automated, the human factor is diminished or eventually removed from automated driving. Until then, a combination of human input and automation is necessary during automated driving. This research focuses on the interaction between humans and machine and how a safe interaction can be designed by incorporating meaningful human control. Initially, the aim was to study how different personalities are reflected in driver workload induced by take-over requests (TORs). However, the COVID-19 circumstances changed the aim to validate the design of the driving simulation experiment by means of an N = 1 experiment. Design variables that have been found to play a role in driver workload are varied in the validation experiment. These variables are the duration of the time budget, traffic density, location of the TOR and task involvement during automated driving. Subsequently, workload was measured by a combination of subjective and physiological indicators and driving performance. Notably, this study includes the Root Mean Square of Successive Differences (RMSSD) and Standard Deviation of Normal to Normal peak intervals (SDNN) as heart rate variability (HRV) measures, which is a novel approach in studies measuring TOR-induced workload. Despite the study design that involved performing an N = 1 driving simulation experiment, significant differences between attribute levels have been found. This study provides recommendations on an empirically-validated set of design variables for future studies involving TORs and driver workload, specifically for the future study on personality and automated driving.

It is generally accepted that the road safety trend of a country is influenced by many factors related to the infrastructure, vehicles, health care etc. Nevertheless, road safety improvement is also a result of a learning process, which can come at an individual and a societal level, according to the current available literature. The former is due to the relationship between exposure (number of events an individual experiences) and risk of road accidents, and the latter is due to the learning process in the society.
The authors argue that the long-term improvement in safety does not only happen through individual and societal (i.e., within society) learning, but also through a third dimension which is the learning process across nations (i.e., in between societies). In this paper we attempt to capture this phenomenon in two ways using data for the EU Member States.
We first analyze countries’ progress in safety improvement in relation to their motorization level. Then we use panel regression to investigate whether the Human Development Index (HDI) as a measure of knowledge is a better predictor of safety instead of exposure measures (like car ownership level). The results show that for many countries lagging behind both in motorization and safety it took less time to converge in terms of safety than in motorization level. We also found that the HDI is overall a better predictor. While a few countries are already getting close to the saturation point in their motorization, an alternative knowledge-based predictor is needed for these countries to better describe trends in mortality rate.

Tram is a sustainable mode of transport, which is able to transform the modern city centers through the various urban regeneration projects that usually come together with it. Tram tracks are often shared with either motorized traffic or pedestrians/cyclists. In this mixed traffic reality, tram accidents are very rare but severe at the same time. Tram driving is really a complex and very demanding task, since the driver should run on time, maintain his/her concentration, predict the behavior of other road users and protect the tram passengers from falls inside the vehicle cabin. In the past, a limited number of studies has attempted to examine tram safety, especially in urban areas, where trams interact with vulnerable road users (VRUs). Subjective notions of traffic safety, that are more connected with the behavior of tram drivers, such as: perceived safety and driving stress, have never been quantified. This thesis covers all the previously mentioned research gaps in order to explain tram safety problems. By utilizing the new knowledge, a list of practical recommendations, which can reinforce tram safety without downgrading system efficiency, is developed. For the development of statistical models related with perceived safety and driving stress, a stated preferences experiment was conducted in Athens. In the survey, tram drivers rated perceived safety and driving stress in a 7-point Likert scale. According to the estimated model of perceived safety, non-exclusive alignments, such as tram/pedestrian malls and mixed traffic operation, downgrade perceived safety. Furthermore, the existence of an unprotected pedestrian crossing and high volumes of VRUs influenced perceived safety negatively. Driving stress is affected mainly by arrival delay and load of standing passengers. Route familiarity is an additional important factor that influences driving stress. The existence of many random beta parameters in perceived safety and driving stress models confirms the subjective nature of these notions. No statistically significant correlation between these two previous notions was observed. Experienced tram drivers believe that they are ready to respond properly in a section that they perceive as unsafe, if they are familiar with it. If there is no familiarity, tram drivers lack confidence and therefore driving stress is increased. Objective safety was examined in the tram network of Amsterdam using accident records and spatial data, such as location of stations and pedestrian/cycling crossings, level of tram lines separation, cycling intensity, city attraction poles and city districts. In Amsterdam, more accidents per km appeared in tram/pedestrian malls; yet, most of the fatal accidents have occurred in tram tracks that are not shared with other road users. High concentrations of tram accidents were also observed around attraction poles and inside the city center, where the flow of VRUs is quite high. Lastly, as a practical recommendation, a consistent design of a tram line using the knowledge from the estimated perceived safety and driving stress models is developed for the first time.

The most common way to evaluate traffic safety is investigating the occurrence and severity of crashes using historical data. This approach however has a number of limitations, the most important of which is probably its reactive nature. An alternative method using non-crash events has gained a lot of attention recently, especially thanks to the rapid improvement of sensing technologies. By gathering trajectory data and calculating various Surrogate Measures of Safety it has become possible to analyse safety without waiting for accidents to happen. Using these indicators combined with Extreme Value Theory (EVT) one can estimate the probability of crashes as extreme (unobserved) events. The primary goal of this thesis is to contribute to the research that has been done so far on the application of Extreme Value Theory to Surrogate Measures for traffic safety analysis. Research questions seek for answers to what we can learn from applying univariate EVT using indicators describing collision course and crossing course interactions, and how we can predict nearness to collision and severity using bivariate EVT models.

By means of a driving simulator experiment and an accident data analysis, this research focuses on two points: (1) an expected difference in turbulence between major fork configurations that differ in the number of mandatory lane changes for rightmost traffic going to the left-diverging roadway, and (2) an expected difference in driver expectancy between left-diverging roadways of a major fork (that are also small-radius connector roads) that differ in the number of traffic lanes they include.

The technology of automated driving systems that assist the human driver are in ongoing development and could potentially improve traffic safety and efficiency. At this moment, a lot of research into automated vehicles is carried out. The City of Amsterdam wants to know what impact AVs can have on traffic safety in their city. Most studies focus on the technology of the vehicle itself and its impact on society. An increasing number of studies is focussing on the human aspects, although most of these researches focus on the driver, while questions remain unanswered on vulnerable road users. At this moment, it is challenging to gain insights in the system of interaction. Due to the still-evolving technology of AVs, the impact on traffic safety cannot be accurately predicted. City of Amsterdam want to start pilots to test AVs on the public roads in order to gain insight in te system of a safe mutual interaction between automated vehicles and vulnerable road users.

First, the system of interaction needs to be known. Using the methodology of Fuzzy Cognitive Mapping (FCM), the determinants and behaviour of the system is identified. FCM is a fairly new method in the field of transport planning, but showed potential for this specific research in which scientific data is limited. The original approach to develop a FCM model is adapted. Therefore, the time it took to develop a conceptual FCM model during a workshop could have been limited and disadvantages of one strategy is balanced or mitigated with the advantages of other stragegies. This research is therefore also assessing if FCM can be a useful method in the field of transport planning.

The FCM model that describes that system of interaction is developed via a literature study and subsequent workshop. This resulted in a model of 21 determinants with 72 connections or relationships. Computations showed that the most important (key) determinants were the following concepts: Safe crossing behaviour, VRU friendly road design, AV friendly road design, Intelligent infrastructure and Identification and recognition. These key determinants, who each describe an idea of something formed by mentally combining all its characteristics or particulars, are considered most important in the system of interaction and should therefore be first be researched in pilots.

The results of the workshop and computations provide a first glance at the system and results. Interviews provided extensive state-of-the-art knowledge on the key determinants. The findings from the interviews are translated into an advice for the City of Amsterdam te develop and execute pilots. These pilots should be able to answer the most important and relevant research questions on the safe mutual interaction between automated vehicles and vulnerable road users in the urban environment.

FCM is found to be useful in the field of transport planning for specific case in which scientific research is limited, with a lack of quantitative data, but available qualitative data from professionals and where human behaviour plays an important role. For the still developing technology of automated driving systems, the method can be useful for as long as quantitative data is not available. As soon as such data is available, other methods are found to be more useful.

Automated Vehicles (AV) will be introduced on public roads in the near future. This would result in automated vehicles sharing the urban space with other road users including drivers of traditional vehicles and vulnerable road users. Pedestrians might be unable to distinguish the vehicle type (traditional or automated) they are interacting with and crossing situations might become confusing, possibly leading to dangerous encounters between pedestrians and vehicles. There is currently little knowledge about the interactions between pedestrians and AVs from the point of view of the pedestrian in a real life environment. The aim of this master thesis is to determine whether pedestrians’ crossing intentions differ when interacting with automated vehicles compared to when interacting with traditional vehicles. An experiment was developed on a closed road where participants encountered a Wizard of Oz automated vehicle and a traditional vehicle in a within-subject design. In the Wizard of Oz set-up, a fake ‘driver’ sat on the driver seat while the vehicle was driven by the passenger with a joystick. Different scenarios were studied regarding vehicle appearance (‘driver’ reading a newspaper, roof signs, hood/side signs) and approach direction (left vs. right). Results showed that the majority of participants reported that the vehicle was (sometimes) driven autonomously, which indicates that the Wizard of Oz was credible. Moreover, most of the participants perceived the differences in vehicle appearance and reported to be influenced by these features. Despite of this, measurements of critical gaps and self-reported level of stress showed no significant differences between the different conditions of vehicle appearance.

Cycling is one of the main transport modes in the Netherlands and the total number of bicycles and the total cycling distances are still rising. Although this seems a positive development, the number of seriously injured cyclists due to bicycle crashes without motor vehicles increased rapidly in recent years. Within these crashes the infrastructure seems to play an important role and research towards the contribution of infrastructure characteristics to bicycle crashes without motor vehicles is needed because this is limited and especially research on more detailed infrastructure characteristics (e.g. bicycle path width) has not been done yet. This study aims to fill in parts of the knowledge gap between the safety problem and the lack of research done on the contribution of infrastructure characteristics. This study was done by using data from a survey held by VeiligheidNL. In total, 3146 cyclists that were treated at an emergency department responded to the survey and provided information on the crash location and the trip origin. A case-control method was applied in this study which compared infrastructure characteristics of case (crash) and control locations (no crash). The control locations were selected from the route of a bicycle crash victim. The design of the method included basic steps for the selection of the controls and four measures to lower the bias in the method and increasing the statistical power of the results. With the application of the case-control method to the data several results were found: First, the width of streets contributes to riding off the road crashes and handlebar collisions, whereas the width of bicycle paths next to bollards contributes to crashes with these objects. Secondly, the presence of intersections decreases the likelihood of kerb collisions. Thirdly, two-directional bicycle paths are more likely to induce bicycle-bicycle crashes and they possibly induce more severe crashes because frontal collisions were found more severe than same direction bicycle-bicycle crashes. And finally, the placement of bollards increases the severity of the injuries caused by the corresponding crashes. These findings could be used to further increase the quality of bicycle design guidelines such that future bicycle infrastructure could be designed safer.