42-nd Summer School of Automatic Control
     Grenoble 2021
Data and Learning for Control
Grenoble 2021
Data and Learning for Control

42nd International Summer School of Automatic Control Grenoble, France 


September, 06-09, 2021


 

Breaks will be done around 10:30 and 15:30



Monday September 6th


8:45-09:00: Paolo Frasca and Mirko Fiacchini: Welcome


9:00-12:30: Mazen Alamir, CNRS, GIPSA-lab, Grenoble


Learning against Uncertainties: In this part, some examples are given where data-driven steps are incorporated in various control-related analysis and design methodologies for nonlinear systems involving uncertain parameters. This includes stochastic Nonlinear Model Predictive Control, stochastic Nonlinear Moving-Horizon Estimation, Data driven adaptation of control updating period and some other topics. The main message is that including data driven blocks in control related design should be a state-of-mind that induces an endless list of possibilities that no course can exhaustively explore.

Slides


14:00-17:30: Melanie Zeilinger, ETH Zurich, CH


Safety Filters for Learning-based Control: While reinforcement learning has shown great success in solving complex control tasks, most techniques cannot ensure safety of the system, in particular during learning iterations. Safety frameworks based on invariance emerged from control theory to address this issue by making use of (approximate) system models. This lecture gives an introduction to predictive control techniques for such safety frameworks, including a hands-on programming session demonstrating the approach.

Slides 1

Slides 2

Coding exercises

Additional material

Solution part 1

Solution part 2



Tuesday September 7th


9:00-12:30: Claudio De Persis, Groningen, NL


Learning Control from Data: Linear and Nonlinear Systems: The lectures focus on a recently introduced approach to design control policies for unknown systems starting from low-complexity input-output data collected during off-line experiments. The approach reduces the design to the solution of data-dependent semidefinite programs, which provide a computationally effective way to deal with the problem of learning control from data. We will see how problems that are central to data-driven control, such as stabilization, optimal regulation and robust controlled set invariance, can be studied with this approach. Some extensions of these results to nonlinear control systems will also be presented.

Slides 1 (for personal use, please do not diffuse)

Slides 2 (for personal use, please do not diffuse)

Additional material


14:00-17:30: Milan Korda, CNRS, LAAS, Toulouse, France


Introduction to the Koopman Operator Approach for Data-driven Analysis and Control of Nonlinear Systems: The Koopman operator approach allows for representing a nonlinear dynamical system by an infinite dimensional linear operator. Finite-dimensional truncations of this operator can then be used to address various control problems such as stability, state estimation and control. This talk will introduce the fundamentals of the approach and how it can be used in a data-driven setting.



Wednesday September 8th


9:00-12:30: Lucian Busoniu, TU Cluj-Napoca, Romania


Reinforcement Learning: Introduction to the Basics and Application to a Communicating Mobile Robot: We describe the class of problems tackled by reinforcement learning (RL) and some essential algorithms. We start with dynamic programming foundations, and build on them to cover some representative model-free, RL approaches. We pay attention to function approximation, which is essential in practice. If time permits, a key deep RL algorithm is explained. Finally, we present an application to a mobile robot that must transmit data over a network with initially unknown transmission rates, which combines model-based and learning updates in an interesting way.

Slides



Thursday September 9th


9:00-12:30: Teodoro Alamo, University of Seville, Spain


Uncertainty Quantification for Kernel Predictive Models: In these lectures, we address the probabilistic error quantification of a general class of kernel prediction methods. We first describe how to develop kernel-based predictive models. We then consider a given prediction model and show how to obtain, through a sample-based approach, a probabilistic upper bound on the absolute value of the prediction error,leading to probabilistic interval predictions. The proposed scheme is based on a probabilistic scaling methodology in which the number of required randomized samples is independent of the complexity of the prediction model. The methodology is extended to address the case in which the probabilistic uncertainty quantification is required to be valid for every member of a finite family of predictors. We illustrate the methodology by means of different case-studies.

Slides

Additional material

Data


14:00-17:30: Fabio Pasqualetti, UC Riverside, USA


Performance and Robustness Guarantees for Data-Driven Control of Complex Networks: Our ability to manipulate the behavior of complex networks depends on the design of efficient control algorithms and, critically, on the availability of an accurate and tractable model of the network dynamics. While the design of control algorithms for network systems has seen notable advances in the past few years, knowledge of the network dynamics is a ubiquitous assumption that is difficult to satisfy in practice. In our work we overcome this limitation, and develop a data-driven framework to control a complex network optimally and without any knowledge of the network dynamics. Our optimal controls are constructed using a finite set of data, where the unknown network is stimulated with arbitrary and possibly random inputs. We will present both centralized methods for the design of optimal controls, as well as distributed algorithms that enjoy favorable convergence and privacy properties. Finally, we will discuss the issue of robustness of data-driven methods to (adversarial) perturbations. We will present methods to quantify robustness as a function of the training data and algorithm's properties, design algorithms that exhibit provable robustness guarantees, and discuss some implications of our results

Slides

Additional material 1

Additional material 2



Friday September 10th


10:00-11:00: Flash presentations track


The participants will be given the opportunity of presenting their own work on topics related to control and learning. The track will be composed by three-minutes-lasting slots, available for the interested particpants under demand.


11:00-12:00: Round table discussion