Maarten Schoukens

Data-Driven Systems and Control

Our aim is to develop a wide range and flexible set of tools and methods for the  data-driven modelling and control of (nonlinear) dynamical systems. This is achieved by combining classical systems and control theory and insights with recently developed deep learning and reinforcement learning approaches. The resulting models and controllers - or policies - aim to be interpretable by the user to promote fast adaption in control engineering applications.

Some questions that arise during this process are:

1. How to design an input to generate data for efficient and robust learning?

2. How can we select a model / controller structure that balances interpretability, flexibility and accuracy?

3. How to train such a model / controller efficiently?

4. How can we validate the model / controller and assess its uncertainty bounds?

These challenges return in many fields of engineering such as motion control problems, structural engineering, telecommunication and high power electronics, fluid dynamics and many more... We pursue both fundamental and applied research in close collaboration with various industrial partners.