How machine learning can transform traditional control methods: At the international 'Learning for Dynamics and Control' conference in Michigan, inIT employee Jörn Tebbe presented a poster on the novel combination of Gaussian processes and model predictive control.
International conference on machine learning and control engineering
The 7th Annual Learning for Dynamics & Control Conference (L4DC) took place at the University of Michigan in Ann Arbor from 4 to 6 June 2025. Researchers from all over the world came together to discuss current developments at the intersection of machine learning and control engineering. A total of 120 scientific papers were presented, including a poster by Jörn Tebbe.
The mathematics and data science working group research assistant, who is led by Prof. Dr. Markus Lange-Hegermann, presented a paper entitled ‘Physics-Informed Gaussian Processes as Linear Model Predictive Controller’, which he co-authored with Andreas Besginow and Markus Lange-Hegermann.
Linking Gaussian processes and model predictive control
In der vorgestellten Arbeit zeigen die Forschenden, wie sich ein klassisches Verfahren der Regelungstechnik – die modellprädiktive Regelung (Model Predictive Control, MPC) – vollständig durch ein Machine-Learning-Modell auf Basis eines Gaußprozesses ersetzen lässt. Anstatt wie bisher eine Optimierung für die Regelung durchzuführen, erfolgt die Steuerung direkt durch eine gezielte Auswertung des Gaußprozesses. Das Verfahren eröffnet somit neue Perspektiven für den Einsatz datengetriebener Methoden in der Regelungstechnik.
‘This method allows known techniques from Gaussian process modelling to be systematically applied to classic control problems,’ explains Jörn Tebbe. ‘This can not only reduce computing time, but also enable the further integration of stochastic concepts into control systems.’
Visibility through international exchange
“Direct exchange with international researchers is incredibly valuable – both for our own work and for the further development of joint methods,’ says Jörn Tebbe. ‘At conferences such as L4DC in particular, you get a good sense of where research currently stands and where it is heading.”
By participating in renowned conferences such as L4DC, inIT can showcase its research on the international stage alongside leading institutions such as MIT, Stanford University, the University of Oxford and ETH Zurich, while also fostering scientific collaboration.