Research opportunities

Master Thesis Project

Understanding Cyclist Experience, Safety, and Satisfaction

Cycling is an integral aspect of sustainable transportation, and promoting cyclist-friendly environments is crucial for enhancing mobility, reducing congestion, and improving public health. However, the quality of cycling infrastructure and the surrounding landscape can significantly impact the experience and safety of cyclists. This project helps in the identification and
understanding of factors that are essential for urban planners and policymakers to design and maintain effective cycling networks.

Case Zuidasdok Amsterdam – Safety impacts of urban design on active transportation

<p><a href="t3://page?uid=52255#1335479">The Zuidasdok project in Amsterdam is an urban revitalization initiative aimed at improving the mobility infrastructure of the Zuidas area. As part of the redevelopment process, local policymakers are eager to understand the safety implications of different urban design elements on active modes of transportation, such as walking, cycling, and other non-motorized forms of travel in this unique, high density urban area.</a></p>

Quantifying the safety perception of AV occupants with respect to the interactions with cyclists and pedestrians

Cyclists and pedestrians, collectively referred to as Vulnerable Road Users (VRUs), including motorcyclists, skateboarders, and equestrians, present a unique challenge to the safety and comfort perception of autonomous vehicle (AV) occupants. The interaction between AVs and VRUs is a key determinant of occupant trust, comfort, and overall well-being. Currently, AV motion planning and decision-making algorithms do not account for AV occupant safety perception, potentially leading to stressful and uncomfortable rides. Recent studies suggest that AVs adopting altruistic behaviors in their interactions with other road users can maximize comfort, trust, and well-being. However, the specific interaction dynamics between AVs and cyclists/pedestrians and their impact on AV occupant safety perception remain unexplored. Furthermore, the quantification and mathematical modeling of
safety perception are lacking. Addressing this knowledge gap can facilitate the development, testing, and optimization of AVs with realistic safety perception algorithms, enhancing passenger well-being and contributing to a safer and more harmonious transportation ecosystem.

Usefulness and necessity of arrow marking in exit lanes, splits and weaving sections

On Dutch motorways, pre-sorting arrows are present in exit lanes, splits and weaving sections (diverging discontinuities). Several neighbouring countries do not apply these markings, and it has become clear that the production and installation of these markings has a relative high negative environmental impact. Furthermore, in car navigation systems might reduce the need for pre sorting markings. Therefore, we are reconsidering the use of pre-sorting arrows on our motorways. However, since these arrows have always been present on our motorways, we assume drivers expect these arrows and anticipate a lane-change using these arrows. The aim of this research is to investigate possible differences in driving behaviour when arrows are not present.

Influence of parallel lines on curve driving

Motorway curves with sharp radii pose a significant challenge for drivers, making them some of the most
dangerous stretches of motorway. Previous eye-tracking research has shown some evidence that incorporating
parallel elements alongside these curves can aid drivers in detecting and navigating them. These elements might
include chevron signs, as well as continuous features like guardrails, noise barriers (as shown in the image
above), or a line of trees. Additionally, it is hypothesised that higher parallel lines offer greater assistance to
drivers in adjusting their driving speed as they approach the curve. This research aims to explore how various
types of parallel lines impact the deceleration process when approaching a curve.

Intensity and diversity of traffic on bicycle tracks and the implications on safety

In 2022 the share of cyclists of total road fatalities increased to about 40%. In recent years, bicycle tracks are being used not only by ordinary bicycles, but also by different types of vehicles, such as pedelecs, light mopeds, and cargo bikes. These different types of vehicles have sometimes large differences in speeds, mass and demand for space requirements. This phenomena increases the heterogeneity on our bicycle tracks, and considering the sustainable safety principles could have negative implications on road safety. However, evidence-based research in this regard is missing. Therefore, there is a knowledge gap on how this increased heterogeneity affects the interactions between road users of different types of vehicles, and the implications on their safety.

How could automated vehicles (AVs) be socially compliant? Insights for social behavior of AVs

How to make the AVs socially accepted by human drivers is vital for the overall traffic safety and efficiency, but remains an unexplored topic. As human drivers possess different social/psychological traits, understanding how humans perceive others’ driving intents/behaviors/styles, and accordingly adapt their own driving behavior under different scenarios can provide valuable insights for developing human-like socially compliant driving behaviors for AVs. The aim of this research is to understand what makes a driving behavior social in different types of interactions by incorporating an online questionnaire and expert interviews. A research intern at external partner is possible.

Deep Reinforcement Learning for developing automated driving model learning from tabula rasa

Deep Reinforcement Learning (DRL) has been applied in various domains and demonstrates its powerful ability. This research aims to try a primary study implementing DRL for selected driving tasks with both lateral and longitudinal control involved (e.g., lane-changing), and learning from tabula rasa. Open-sourced microscopic traffic simulation platforms (e.g., SUMO), as well as Automated Driving System simulation tools (e.g., CARLA) are all available. The integrated tool, OpenCDA, which combine SUMO+CARLA, could also be a candidate platform for simulating and training the AV model. A research intern at TNO is possible.

MSc Thesis: Deep learning for abnormal driving behaviour detection

Accurate detection of abnormal driving is vital and prerequisite for imitate training of a human-like driving model. Existing abnormal driving detectors are mainly based upon shallow supervised learning, which require abundant labelled data. Usually, there is plenty of normal driving data available, while collecting data involving abnormal driving is difficult and dangerous. How to make use of available data to train a model able to detect potential upcoming abnormal samples would be a meaningful and challenging task. The aim of this research is to explore possible semi-supervised deep learning methods to identify abnormal driving behaviours using open-sourced datasets. Pipeline is available and a research intern at TNO is possible.

What are the potential road safety benefits of ADAS deployment in LMICs and what are the main barriers?

The development and deployment of Adaptive Driver Assistant Systems in motorized vehicles promises to improve traffic safety by gradually increasing the guidance role of vehicles. In-vehicle support may help drivers to behave safely in today’s complex traffic situations. These technologies could therefore have a great potential of improving traffic safety in low and middle income countries (LMIC), where actually the majority of road fatalities and serious injuries occur worldwide. Despite the criticality of the situation, there is a clear knowledge gap regarding the potential road safety benefits of ADAS deployment in LMICs and the main barriers for large scale deployment and utilization of the vast developments of technologies that can improve traffic safety.

MSc Thesis: Robust Lane Detection using Image Sequential Attention Based Transformer Model

Most available methods focus on detecting the lane from one single image, and often lead to unsatisfactory performance in handling some extremely-challenging situations. As lanes are continuous line structures, the lane that cannot be accurately detected in one single frame may potentially be inferred out by incorporating information of previous sequence of images. The aim of this research is to develop a sequential DL model incorporating Vision Transformer with continuous image frames as inputs to detect lanes in the last frame. The research group had already built the pipeline and collected a large scale driving scene data in the Netherlands for testing the model’s robustness.

Share this page: