News & updates from the group

ACE AI & Data Talk

Data-driven models in predictive control: Uncertainty quantification and robust designs

Talk in the ACE AI & Data Talk series on data-driven approaches in predictive control, focusing on uncertainty quantification and robust design methods.

Recording available online:

New preprint

Certainty-equivalent adaptive MPC for uncertain nonlinear systems

Johannes Köhler

We developed an adaptive model predictive control (MPC) scheme with online learning for time-varying systems. The method enables online adaptation while maintaining robustness and guarantees on tracking performance and constraint satisfaction.

New preprint

Finite-sample bounds for multi-output system identification
Léo Simpson, Katrin Baumgärtner, Johannes Köhler, Moritz Diehl

The paper derives finite-sample confidence bounds for linear regression in the multi-output setting. The results extend self-normalized martingale bounds beyond scalar outputs and provide tighter confidence sets for system identification.

New L-CSS paper

Exponential stability of data-driven nonlinear MPC using input/output models
 Lea Bold, Irene Schimperna, Karl Worthmann, and Johannes Köhler, IEEE Control Systems Letters (L-CSS)

The work shows that learned surrogate models can enable model predictive control with exponential stability guarantees for unknown nonlinear systems.

New L-CSS paper

Optimal uncertainty bounds for multivariate kernel regression under bounded noise: A Gaussian process-based dual function
A Lahr, A Scampicchio, J Köhler, MN Zeilinger
Optimal error bounds for non-parametric regression under bounded noise are formulated as a Gaussian Process using duality
 

Our group at European Control Conference (ECC) 2026

Tutorial session
Safe-by-design control using robust MPC
Slides available online:

Workshop presentations
Stochastic and data-driven MPC

Invited session (organiser)
Advances in MPC: safe decision-making under uncertainty

Conference talk
MPC with reduced-order models

 

New preprint

A model-free approach to control barrier functions for higher-order systems
L Lanza, J Köhler, D Dennstädt, T Berger, K Worthmann
 
Control barrier functions (CBFs) are designed without a model using concepts from funnel control and appropriate modifications for higher-order systems.
 
 

 

New Automatica paper

Finite-sample-based reachability for safe control with Gaussian Process dynamics
Manish Prajapat, Johannes Köhler, Amon Lahr, Andreas Krause, Melanie N. Zeilinger

This work provides a method to ensure safe operation using Gaussian process models in a model predictive control framework by over-approximating the reachable set based on predictions from a finite number of GP samples.

 

New Automatica paper


A robust and adaptive MPC formulation for Gaussian process models
Mathieu Dubied, Amon Lahr, Melanie N. Zeilinger, Johannes Köhler
This paper provides a predictive controller that ensures robust constraint satisfaction based on Gaussian process models that are adaptive using online data. 
 

New EJC paper

Robust reduced-order model predictive control using peak-to-peak analysis of filtered signals
Johannes Kohler, Carlo Scholz, Melanie Zeilinger
European Journal of Control (EJC)
This work presents a predictive control method for reduced-order models that retains safety guarantees and reduces conservatism by over four orders of magnitude compared to existing approaches.
 

New TAC paper

Stochastic MPC with Online-optimized Policies and Closed-loop Guarantees

Marcell Bartos, Alexandre Didier, Jerome Sieber, Johannes Köhler, and Melanie N. Zeilinger

IEEE Transactions on Automatic Control (TAC)

The work develops a stochastic MPC framework that optimizes over closed-loop policies to minimize expected cost while providing guarantees for probabilistic constraint satisfaction in infinite-horizon operation.

New TCST paper

Robust Convex Model Predictive Control With Collision Avoidance Guarantees for Robot Manipulators
Bernhard Wullt, Johannes Köhler, Per Mattsson, Mikael Norrlöf, and Thomas B. Schön,

IEEE Transactions on Control Systems Technology (TCST)The work presents a robust motion planning for robot manipulators, enabling real-time re-planning with rigorous  safety guarantees despite model uncertainties.