Data-driven methods

Several years ago our main challenge as researchers was to gather enough data, and the research methods were limited. Nowadays, our main challenges are to understand how to analyse and handle big data, and how to select suitable research methods. The advancement in technology and simulation capabilities opens up the door for a vast research methodologies and possibilities, including virtual reality, augmented reality, simulators, and intelligent vehicles as probe vehicles collecting data in real-time. How can we benefit from all this to analyse safety? And how can we integrate different research methods to increase the validity of our results? 

Relevant research:


Project title


Yongqi Dong

(PhD candidate)

Data-driven research for expanding AVs’ operational design domain in mixed traffic

Both data-driven deep learning models and state-of-the-art deep reinforcement learning models are being explored to develop reliable algorithms and policies for AVs’ interacting with infrastructure and other road vehicles on different roads and under different traffic conditions to improve the safety, efficiency and social compliance of AVs in mixed traffic. Various open-sourced datasets (e.g., Audi A2D2, highD, and INTERACTION Dataset) together with self-collected (filed test) data will be examined and explored to train and validate the models.

Nagarjun Reddy

(PhD candidate)

Studying human drivers’ behaviour in mixed (HDV and AV) traffic environments

Studying human driving behaviour in a future scenario where AVs are also present on roads is challenging, not least due to this being a future scenario. Investigating this requires strong empirical underpinning, therefore the use of different data collection techniques. Methods such as driving simulator experiments, controlled field tests, and naturalistic driving data sets will be explored to this end.

/* */