A one-day workshop called Group of Global Excellence (GoGE) was held on October 25. This workshop was organized by CDSL and supported by Brain Korea Program of the Korean government
Professor Florian Dörfler of the Department of Information Technology and Electrical Engineering at ETH Zürich, Switzerland, was invited as the main speaker for this workshop. Prof. Dörfler delivered a seminar under the title of “Data-Driven Control Learning Pipelines: general perspective & recent results for LQR”. In this seminar, he explored the evolution of data-driven control theory, providing insights into its historical context and theoretical foundations. He also suggested the pipelines, limitations, and goals of data-driven control theory which is represented as model-free, adaptive, robust, cheap, and tractable. Under these considerations, he presented his recent works of model-free and online data-driven LQR. He proposed novel parameterization and regularization techniques for model-free LQR and introduced policy gradient methods for adaptive LQR. He showed that proposed controller not only has some theoretical certificates, but also works well on complex systems.
Several students of CDSL participated as speakers in this workshop. Deuksun Hong shared his recent result under the title “Designing Reinforcement Learning Input Based on Nonlinear System Obeservability”, Yeongjun Jang shared “Path Generating Inverse Gaussian Process Regression for Data-driven Ultimate Boundedness Control of Nonlinear Systems”, and Joowon Lee shared “Stability Analysis on Data-driven Simulation and Inversion”. Following each presentation, Prof. Dörfler actively engaged with the speakers, posing insightful questions and fostering productive discussions.
The workshop was highly enriching, and we sincerely appreciate Prof. Dörfler’s enthusiastic presentation and active participation.
You can find his presentation slide at here (LQR Learning Pipelines).