TITLE

Data-driven modelling, analysis, and control using the Koopman operator

LENGTH

Full-day workshop from 8:40 am to 5:30 pm.

ORGANIZERS

  • Petar Bevanda
    (TU of Munich, GER)
  • Sandra Hirche
    (TU of Munich, GER)
  • Armin Lederer
    (ETH Zurich, SUI)
  • Alexandre Mauroy
    (University of Namur, BEL)
  • Karl Worthmann
    (TU Ilmenau, GER)

CORRESPONDENCE

LINKS

CDC website

ABSTRACT

The linearity of Koopman operators and the forecast simplicity of linear time-invariant (LTI) models coming from their functional representation in a space of "observables" lead to their increased popularity for learning dynamical systems. This representational simplicity inspired a bevy of system identification approaches and holds promise for solving many classes of nonlinear control problems through lifting nonlinear systems into suitable spaces of observables. Given that operator-theoretic system identification methods are still under active development, it is critical to present current results, main challenges and point to fruitful directions for making Koopman-based approaches more mature for systems-and-control applications.

This workshop aims to present a broad overview of state-of-the-art Koopman operator approaches from the perspective of systems and control theory. The main objective is to provide a multifaceted perspective through an introduction to fundamental concepts, current opportunities/challenges, and recent advancements – striking a balance between a tutorial style and presenting the latest advancements.

     

PROSPECTIVE AUDIENCE

The workshop targets a broad audience ranging from graduate students and researchers looking for an introduction to a new and active area of research, to practitioners interested in data-driven design methods. The required background is basic familiarity with systems and control. As the talks cover a variety of relevant and modern topics, this workshop provides an excellent overview of the state-of-the-art in data-driven control. 

 

 

 

 

 

 

 

 

 

 

STRUCTURE AND PROGRAM

The workshop is organized around the following four thematic areas:

  • A tutorial introduction, (im)possibilities and dynamic mode decomposition (DMD)
  • Choosing observables: from exact embeddings, and kernel methods to representation learning
  • Perspectives on optimal, predictive, and robust control
  • Certifiable learning and control design
     

The workshop consists of 12 talks by experts in the field. The workshop is planned as a full-day event with the schedule as depicted in the following table. The first two presentations in the morning will include a tutorial style introduction to the topic, such that the overall workshop becomes easily accessible for an audience with a diverse background.

 


 

 

Time schedule with (tentative) titles of the presentations

 
TimeslotPresenter(s)Topic
Morning Session:
8:40 amSandra HircheWelcome and opening remarks
8:50 amAlexandre MauroyA tutorial introduction to Koopman operator theory
9:30 amZlatko DrmačAdvances in DMD - using the residuals and the Koopman-Schur decomposition
10:00 amCoffee break
10:30 amMatthew ColbrookSpectra of Koopman operators: what is computationally (im)possible?
11:00 amShankar A. DekaInterpretable and verifiable learning for Lyapunov-based certificates
11:30 amNecmiye OzayProperties of immersions for systems with multiple limit sets with implications to learning koopman embeddings 
12:00 pmLunch break
Afternoon Session:
1:30 pmKarl WorthmannControl of nonlinear systems using extended dynamic mode decomposition (EDMD) with closed-loop guarantees
2:10 pmJoão HespanhaOptimal control of switched Koopman models
2:40 pmRoland TóthExact Koopman representation of nonlinear systems: from control inputs to finite dimensional embeddings
3:10 pmCoffee break
3:40 pmPetar Bevanda, Armin LedererKoopman kernels for learning dynamical systems
4:20 pmHaruhiko Harry AsadaControl coherent Koopman modeling: extending Koopman operator to control problems without approximation
4:50 pmBoris HouskaData-driven stochastic control: exploring a PDEconstrained optimization perspective
5:20-5:30 pmSandra HircheClosing remarks
 

Abstracts