Strong DSME Presence at Top Control Conference
Four research papers and four invited sessions have been accepted at the IEEE Conference on Decision and Control (CDC) to be held December 6th to 9th, 2022 in Canún, Mexico, marking a strong presence of the Insitute for Data Science in Mechanical Engineering (DSME). The CDC is recognized as the leading international conference on automatic control.
Here is a short overview of the accepted papers.
Towards Remote Fault Detection by Analyzing Communication Priorities
In this paper, we build upon a recently proposed predictive triggering architecture that involves communication priorities shared throughout the network to manage limited bandwidth. We propose a fault detection method that uses these priorities to detect errors in other agents and demonstrate it in hardware experiments.
click here for more details (arXiv)
Improving the Performance of Robust Control through Event-Triggered Learning
We propose an event-triggered learning algorithm that decides when to learn in the face of uncertainty in the LQR problem with rare or slow changes. Our key idea is to switch between robust and learned controllers to improve performance which we demonstrate in a numerical example.
On Controller Tuning Using Time-Varying Bayesian Optimization
This paper considers Bayesian optimzation for controller tuning in time-varying envirnments. We incorporate suitable assumption for this controller tuning problem in the Gaussian process surrogate model such as prior knowledge of a convex objective function and demonstrate that leveraging these assumtions significantly ourperforms state-of-the-art approaches.
Learning Functions and Uncertainty Sets Using Geometrically Constrained Kernel Regression
In the paper, we propose to combine geometric constraints, expressing prior knowledge, and kernel methods to learn a nominal prediction as well as uncertainty sets for an unknown target function from data. This problem is very relevant for learning-based control, and our approach allows us to circumvent problematic and unrealistic assumptions, contributing to making learning-based control approaches safer and more practical.
Four Invited Sessions on "Learning-based Control"
Besides the four mentioned papers, also four invited sessions on the research topic of "Learning-based Control" have been accepted at the CDC. Here, scientists from top international universities and research labs will be presenting their latest research at the intersection of machine learning and automatic control. The sessions are co-organized by Sebastian Trimpe (RWTH Aachen University), Angela Schöllig (Technical University of Munich, Univ. of Toronto), Melanie Zeilinger (ETH Zürich), and Matthias Müller (Univ. Hannover).
Learning-based Control has become the "hot topic" at CDC
Because of the relevance of data-based techniques and machine learning in the area of automatic control, Sebastian Trimpe had initiated and co-organized the first edition of invited sessions on "Learning-based Control" at CDC already in 2016. Starting with a single session in 2016, the sessions have been held every year and become increasingly popular at the conference. In recent years, the sessions were among the most popular of the entire conference with participants lining up in front of the conference rooms to listen to the talks. This year, there will be four sessions with a total of 24 presentations from top research labs around the world.
Information on the sessions of previous years is summarized at Invited Session Series on Learning-based Control.