Research opportunities

Master Thesis Project

The Role of Transition Control in Enhancing Traffic Flow and Safety

In certain situations when the driving conditions are outside the operational design domain of the automated driving system, the driver would need to take back control and drive the vehicle manually. These transitions of control might have negative implications on the traffic performance. In this direction, in this thesis you will investigate the impact of transition control on traffic flow safety and efficiency.

Acceptability and willingness-to-pay for AV: exploiting immersive data collection techniques to improve accuracy of results

This project aims to address the challenge of reducing the hypothetical bias to derive more accurate and reliable policy measures, such as willingness to pay and acceptability of AVs with certain characteristics. This will be achieved by (1) focusing on specific aspects of automation, so that potential respondents are not asked to imagine an entirely new mode and (2) using driving simulator scenarios so that respondents can practically experience such automation features. In particular, this study will investigate the reliability of different types of data collection methods when it comes to different features of automation.

MSc Thesis: 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.

Interaction of pedestrians with automated transport in Virtual Reality environment

This proposed Master thesis topic is related to the SIPCAT project. SIPCAT stands for Safe Interaction of pedestrians and cyclists with automated transport. Automated vehicles can contribute to sustainable and efficient transport. When driving slowly, cautiously and predictably, they could share the road and interact with vulnerable road users. Conducting real-life experiments is very expensive and might put the road users in real risk, and therefore, it is also unethical. In this project we study the traffic safety of this interaction within the 3D VR Digital Twin of the Marineterrein Amsterdam Living Lab (MALL).

MSc Thesis: 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.

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.

MSc Thesis: 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.

Deel deze pagina: