Abstract
Targeted exploration and optimal experiment design have a long-standing role in system identification and control, offering systematic methods to design excitation inputs that improve parameter estimation. However, classical approaches typically lack guarantees of excitation on the exploration inputs. In this talk, we present a unified approach to estimate parameters of linear systems subject to either stochastic or non-stochastic disturbances. We provide a priori, non-asymptotic guarantees on the parameter estimation error that can be ensured before exploration. Furthermore, we are interested in capturing the 'dual effect' - the interplay between control performance and exploration. Leveraging tools from gain-scheduling, we integrate the design of targeted exploration and robust control, resulting in a tractable dual control strategy for linear systems.
Biographical Information
Janani Venkatasubramanian is a Ph.D. candidate at the Institute for Systems Theory and Automatic Control, University of Stuttgart, advised by Prof. Frank Allgöwer. She is also a member of the International Max Planck Research School for Intelligent Systems (IMPRS-IS). She received her M.Sc. in Electrical Engineering from Delft University of Technology in 2018. In 2023, she visited the University of Oxford for a research stay, hosted by Prof. Mark Cannon. Her research focuses on the intersection of system identification and robust control, with particular interests in optimal experiment design and dual control.