Three Research Lines
DeepTraffic brings together three complementary research lines to improve how traffic is predicted and managed: Learn, Adapt, and Explain. We combine knowledge of how traffic behaves with AI and real-time data, and focus not only on better predictions but also on making them useful and understandable for traffic professionals.
Together, Learn, Adapt, and Explain form an integrated approach: from modelling traffic dynamics, to continuously updating predictions, to supporting informed decisions under uncertainty.
1. Learn
Learning how traffic behaves
We develop hybrid traffic models that combine traffic flow theory with machine learning. Techniques such as Physics-Informed Neural Networks, Neural ODEs and residual learning will be explored to capture traffic dynamics while respecting fundamental physical principles.
In short: We let AI learn where traffic theory falls short—without throwing traffic theory away.
2. Adapt
Keeping predictions in sync with reality
We develop AI-augmented data assimilation methods that combine real-time traffic observations with traffic models. This includes estimating traffic states and changing demand patterns, as well as predicting the conditions entering and leaving parts of the network. The aim is to make predictions continuously responsive to the latest available information.
In short: We use every relevant piece of information to keep predictions up to date.without throwing traffic theory away.
3. Explain
Making predictions understandable
DeepTraffic focuses not only on predicting traffic, but also on explaining the mechanisms behind those predictions. We investigate ways to visualise uncertainty, compare alternative scenarios and identify possible causes and consequences of traffic developments.
The aim is to support traffic professionals in making informed decisions under uncertainty.
In short: Not just “what will happen?”, but also “why?”, “how certain are we?” and “what if we intervene?”