Interaction between road users is a fundamental part of the traffic system. The advent of automated vehicles (AVs) has given rise to requirements for interactions between AVs and other road users, expressed in high-level terms like ‘demonstrate anticipatory behaviour’, ‘not confusing other road users’, and ‘being predictable and manageable for other road users’. Operationalizing these social driving behaviours requires social science knowledge on human interaction. However, translating social driving behaviour requirements unambiguously to the engineering domain necessitates that social scientists have a rudimentary understanding of the language of engineers (and vice versa). The present study seeks to accommodate interdisciplinary collaboration between social scientists and engineers by providing insight into current AV technological capabilities with regards to social driving behaviour and road safety, and their development in the near future. To this end, an exploratory interview study was performed with 7 engineers with backgrounds in industry, academia, research institutes, and/or vehicle authorities. The engineers provided several real-world examples of implications of AV algorithms on social driving behaviour. Thematic analysis of the interview transcripts resulted in clusters of themes relating to the product development process: requirements (i.e., societal, legal, customers), development (i.e., process, implementation), and evaluation (i.e., assessment, monitoring). Choices made in each of these phases appear to influence the final behaviour of automated vehicles in traffic. Knowledge on social driving behaviour and its impact on traffic safety can guide these choices to ensure safe operation of AVs within the social environment of traffic.
Before the achievement of fully automated driving, drivers are still required to take control of the vehicle when necessary. The performance of this takeover process is influenced by various factors, such as the driver's state of alertness and the characteristics of the traffic environment. This study explored how obstacle characteristics in the traffic scenario, driver alertness, and non-driving related tasks (NDRTs) affected the performance of the takeover process, particularly the situation understanding time and takeover reaction time, in conditionally automated driving. The AdVitam dataset published by Meteier et al. (2023) was used in this study, where a driving simulation experiment was conducted with 90 participants, collecting data on electrodermal activity (EDA) and subjective perceived risk in six different scenarios (Deer, Cone, Frog, Can, False Alarm 1, and False Alarm 2). A Structural Equation Model (SEM) was constructed to investigate the causal relationship among obstacle movability, obstacle inherent hazard, NDRTs, driver perceived risk, alertness prior to the takeover request, and takeover process. The results indicated that driver alertness, measured by features extracted from EDA, played a significant role in takeover reaction time. Higher alertness leaded to quicker reactions when taking over control. Furthermore, perceived risk, influenced by obstacle movability and inherent hazard, significantly mediated the relationship between obstacle characteristics and takeover reaction time. Additionally, obstacle movability affected situation understanding time directly. These findings suggest that obstacle characteristics and driver physiological signals can be combined for an accurate prediction of drivers' situation understanding time and takeover reaction time in automated vehicles, thereby enabling adaptive adjustment of takeover warning lead time and enhancing human-machine interaction experience during automated-to-manual transition.
Effective hazard perception training is crucial for improving road safety, yet the factors influencing its efficacy remain underexplored. This study assessed the efficacy of an interactive hazard perception training program utilizing Virtual Reality (VR) technology and the Unity 3D Platform. A driving simulator experiment was conducted with 34 licensed drivers, who were exposed to 20 high-risk urban driving scenarios. Data were collected over three test sessions: baseline, immediately post-training, and 7 days post-training. The study focused on two types of conflict scenarios-horizontal and vertical. Our analysis showed that drivers' HPTs followed a Weibull distribution, which allowed us to develop an accelerated failure time (AFT) model. The results indicated that scene conflict type and driver age positively affected HPT, while test session and initial driving speed showed an inverse relationship. Importantly, the training significantly improved drivers' ability to identify hazards, with notable improvements in both horizontal and vertical conflict scenarios observed immediately after training and 7 days later. This study demonstrates that VR-based hazard perception training effectively enhances drivers' hazard detection skills. These findings contribute to the development of more standardized and effective models for hazard perception training, offering potential for wider application in driver education and road safety programs.
Traffic accidents are primarily caused by human factors, with insufficient risk perception ability emerging as a critical contributor to decision-making errors. In light of this, this study aims to construct a comprehensive evaluation index and predictive model for drivers' risk perception ability, thereby providing methodological support for addressing challenges in risk perception assessment and dynamic intervention in risk prevention. First, multimodal data-including eye movement, electrocardiogram, and behavioral indicators-were collected from 80 drivers in a driving simulator under explicit and latent risk scenarios. Second, a risk perception ability index integrating six key indicators was developed using Principal Component Analysis, achieving a cumulative variance contribution rate of 88.48 %. Validation using the Dynamic Time Warping algorithm demonstrated that the mean minimum distance between the index and each base indicator was 64.75, indicating its effectiveness in comprehensively representing multi-dimensional features. Third, the Extreme Gradient Boosting model was employed to predict the risk perception ability index. Results showed an adjusted R2 of 0.834, significantly outperforming benchmark models such as Random Forest. Analysis using Shapley Additive Explanations revealed that external environmental variables (e.g., training content and risk type) had the most substantial impact on the index. Cognitive ability positively affected individual traits, while an aggressive driving style had a negative influence. Additionally, the average risk perception ability index of elderly drivers was lower than that of younger drivers. This study achieves a quantitative and dynamic evaluation of risk perception ability, offering valuable support for developing and optimizing driver assessment, training programs, and proactive risk intervention systems.
Key Performance Indicators (KPIs) are an important instrument for road safety policy making. This paper presents a KPI for safe urban roads that was developed and tested in the European Trendline project. Based on discussions with 18 experts from 8 countries, it was decided to start with the basic indicator 'Share of 30 km/h road length of the total length of urban roads' which was tested in five pilot countries using available local, national and international (GIS) databases. The share of 30 km/h roads appeared to differ considerably between the pilot countries, ranging from less than 1% in the city of Silistra in Bulgaria to 73% in the Netherlands. The pilots showed that it is possible to calculate the KPI, although it is important to check the quality of the speed limit data and there will be some differences in selections due to data availability and local context. Also the use of additional indicators related to the concept of safe speeds was explored in the pilots. The calculation of these indicators appeared to be more challenging and further research is recommended to further develop these additional KPIs.
