May 4, 2018

Talk of Dipl.-Ing. Matthias Lorenzen

--- Title: On Finite Sample Approximations for Predictive Control under Uncertainty

May 4, 2018

Dipl.-Ing. Matthias Lorenzen
Institute for Systems Theory and Automatic Control
University of Stuttgart
Stuttgart, Germany


Friday 2018-05-0413:00
IST-Seminar-Room 2.255 - Pfaffenwaldring 9 - Campus Stuttgart-Vaihingen




Motivated by results in statistical learning theory, decision making based on observed data has gained an increasing attention in the control community. If sampling-based methods are used in predictive control, interesting questions regarding relevant properties like feasibility of the online optimization, constraint satisfaction, and closed-loop stability arise.

In this talk, we present an MPC framework based on stochastic models and discuss the theoretical foundation as well as main differences to nominal or robust MPC. Subsequently, finite sample approximations of the online stochastic program are presented which lead to computationally tractable algorithms. We discuss the difference between online and offline sampling, analyze the resulting closed loop, and prove relevant system theoretic properties. Finally, we briefly touch upon experimental results of the developed theory applied to robust guidance and control strategies for automated rendezvous operations between spacecrafts.





Biographical Information



Matthias Lorenzen received his diploma degree (Dipl.-Ing.) in Engineering Cybernetics from the University of Stuttgart in 2013 and is currently a doctoral student within the Graduate School of Simulation Technology. During his studies he spent one year as a DAAD Fellow at Harvard University and four months at Daimler AG for an internship in the research department Environment Perception. In 2013, he joined the Institute for Systems Theory and Automatic Control where he is a research and teaching assistant. In 2016, he spent three months as visiting researcher at the University of Oxford. His doctoral studies are focused on model predictive control under uncertainty.


To the top of the page