A Full-Day Pre-Congress Workshop for 2026 IFAC World Congress
Learning from Data: Principles, Methods, and Emerging Ideas & Applications in Iterative Learning Control
Sunday, 23 August 2026
Busan, Republic of Korea
Organizers:
Professor Ying Tan, University of Melbourne, Australia
Professor Tom Oomen, Eindhoven University of Technology, The Netherlands
Professor Kevin L. Moore, Colorado School of Mines, USA
Professor Bing Chu, University of Southampton, UK
Professor Kira Barton, University of Michigan, USA
Dr Leontine Aarnoudse, Norwegian University of Science and Technology, Norway
Workshop Outline:
Data-driven learning control has emerged as a powerful framework for achieving high-performance operation in complex, uncertain, and repetitive systems. By exploiting historical data, system models, and online measurements, such controllers can adapt, optimise, and progressively improve performance over time. Iterative learning control (ILC) and related approaches have been successfully applied across a wide range of domains, including robotics, precision manufacturing, autonomous systems, chemical process control, and rehabilitation, particularly in settings where accurate system models are difficult to obtain and the tasks are inherently repetitive.
This full-day workshop aims to provide participants with a comprehensive overview of the principles, design methods, and emerging applications in data-driven learning control. The workshop will:
Introduce the fundamentals and historical development of iterative learning control and data-driven control.
Present state-of-the-art methodologies, including optimization-based design, frequency-domain approaches, norm-optimal discrete repetitive processes, learnability, and persistence of excitation.
Cover spatial and distributed ILC techniques and their practical implementations.
Showcase real-world applications through case studies and industrial examples.
Provide dedicated discussion sessions for open challenges, emerging directions, and potential collaborations.
By the end of the workshop, participants will have a clear understanding of data-driven learning control frameworks, the ability to connect theoretical concepts with practical applications, and insight into future research opportunities in this rapidly evolving field.
Desired Learning Outcomes:
By the end of the workshop, the audience will be able to:
Identify (traditional and emerging) applications of learning control
Design learning control laws using state-of-the-art methodologies
Understand the fundamentals and historical development of data-driven learning control
Describe and apply state-of-the-art design methods, including optimization-based approaches, frequency-domain design, norm-optimal discrete repetitive processes, and spatial/distributed learning control.
Connect theoretical concepts to practical applications through case studies and industrial examples.
Identify emerging challenges and research directions in data-driven learning control.
Intended Audience and Desired Prerequisite:
Academics and researchers working in control systems, robotics, mechatronics, and automation.
Postgraduate students interested in data-driven and learning-based control methodologies.
Industrial engineers and practitioners involved in advanced control, precision systems, and autonomous or intelligent systems.
In recent years, the workshop organizers have been actively involved in organizing invited sessions related to learning-based and repetitive control at major international conferences, including the American Control Conference (ACC), the IEEE Conference on Decision and Control (CDC), and the IFAC World Congress 2017. These sessions have consistently been well attended, typically attracting 35–40 participants, with balanced participation from North America, Europe, and Asia, demonstrating sustained community interest and broad international appeal.
A related workshop was presented at the IFAC World Congress 2017, where the primary emphasis was on recent theoretical developments. In contrast, the proposed IFAC 2026 workshop adopts a distinctly different and more application-oriented focus. While fundamental principles and dominant design frameworks of data driven learning control are introduced, the workshop places particular emphasis on practical implementations, case studies, and real-world deployment challenges.
By foregrounding applications and practical insights, the IFAC 2026 workshop aims to broaden its appeal to both academic researchers and practitioners, and to demonstrate how data-driven learning control methods are being translated into impactful engineering solutions.
Speakers:
Professor Ying Tan, University of Melbourne, Australia
Professor Tom Oomen, Eindhoven University of Technology, The Netherlands
Professor Kevin L. Moore, Colorado School of Mines, USA
Dr Leontine Aarnoudse, Norwegian University of Science and Technology, Norway
Workshop Schedule:
09:00-09:05 Opening – Welcome, workshop overview, and motivation (Ying Tan)
09:05-09:50 Introduction to ILC and Frequency-Domain Design
Speaker: Tom Oomen, Eindhoven University of Technology, The Netherlands
Description: Iterative Learning Control (ILC) and Repetitive Control (RC) can compensate for repeating disturbances. The aim of this part is to develop a basic ILC algorithm. A frequency domain viewpoint is taken for the analysis and design. Examples show how such an approach is particularly suitable for applications where frequency domain system models are used. Several implementation aspects, including noncausal designs, are highlighted. The focus is on ILC design, and it is pointed out how it relates to a very similar RC framework.
09:50-10:20 Classical ILC/DRP Design from an Algebraic (Matrix-Fraction) Perspective – Part 1: Framework
Speaker: Kevin L. Moore, Colorado School of Mines, Golden CO US
Description: In this two-part talk we introduce and summarize the classical results for controller design when systems operate in an iterative, or repetitive, manner, with both a time-domain axis (along-the-pass) and on an iteration-domain axis (pass-to-pass). In Part 1, focusing on discrete-time systems, we introduce a one-step iteration delay operator, analogous to the classical one-step time delay operator. With this concept we define an algebraic, matrix-fraction framework from which we can simultaneously design controllers that operate both along-the-pass and from pass-to-pass.
10:20-10:40 Morning Coffee Break
10:40-11:10 Classical ILC/DRP Design from an Algebraic (Matrix-Fraction) Perspective – Part 2: Design
Speaker: Kevin L. Moore, Colorado School of Mines, Golden CO US
Description: In Part 2 of this two-part talk, we illustrate how the algebraic, matrix-fraction framework introduced in Part 1 enables a variety of classical time-domain control concepts, such as the internal model principle, MPC, Kalman filtering, PID design, monotonic convergence, l1, H2, H∞, and more, to be applied in the iteration-domain. We discuss several of these ideas in detail with examples, including PID design with current iteration tracking error (CITE) for monotonic convergence and internal model principle-based design.
11:10-12:00 Norm-Optimal ILC
Speaker: Tom Oomen, Eindhoven University of Technology, The Netherlands
Description: An optimization-based approach to designing ILC controllers is presented. This allows a systematic design approach, and initially an approach based on linear algebra is developed. This allows to systematically analyse different notions of convergence, including monotonicity, as well as time-varying solutions. Design considerations are outlined, as well as relations to Riccati-based solutions.
12:00-13:30 Lunch Break
13:30-14:20 Learnability and Persistence of Excitation
Speaker: Ying Tan, University of Melbourne, Australia
Description: In this talk, we introduce the concepts of learnability and persistence of excitation (PE) as fundamental principles underlying iterative learning, adaptive control, and human motor learning. We show how persistence of excitation governs convergence speed, parameter identification, and learning performance, and discuss why good tracking does not necessarily imply successful learning. Using examples from iterative learning control and human motor learning, we illustrate how feedback, feedforward adaptation, and excitation interact during the learning process. The resulting framework provides practical insights into designing learning systems that achieve faster convergence, improved robustness, and enhanced learning efficiency.
14:20-14:50 Nonlinear ILC for Disturbance Discrimination
Speaker: Leontine Aarnoudse, Norwegian University of Science and Technology
Description: Including a nonlinearity in a controller can increase the design freedom significantly. In this talk, we show how this idea can be applied to norm-optimal and frequency-domain ILC, develop convergence conditions that allow for intuitive tuning, and use the additional design freedom to differentiate between repeating and non-repeating disturbances. This leads to fast and accurate attenuation of repeating disturbances, without amplifying iteration-varying disturbances and without sacrificing the convergence speed.
14:50-15:20 Fast-convergence ILC
Speaker: Ying Tan, University of Melbourne, Australia
Description: In this talk, we introduce fast-convergence iterative learning control (ILC) algorithms based on modified rescaled gradient descent. We show how optimisation techniques can be incorporated into the ILC framework to significantly accelerate convergence while maintaining robustness under practical engineering conditions. Unlike existing rescaled gradient methods that rely on restrictive assumptions, the proposed approach relaxes these requirements and provides practical design procedures for implementation. We discuss the trade-off between convergence speed and learning accuracy, present convergence properties, and illustrate the effectiveness of the proposed algorithms through representative examples and applications.
15:20-15:40 Afternoon Coffee Break
15:40-16:10 Data-Driven ILC (continuous), Norm-Optimal ILC (discrete), and DRP
Speaker: Kevin L. Moore, Colorado School of Mines, Golden CO US
Description: Recently, Willem’s Fundamental Lemma has been used to develop what can be considered to be model-free, data-driven control strategies. In this talk we show how to apply this idea to ILC and DRP controller design, including for norm-optimal ILC, thus producing truly model-free ILC/DRP algorithms.
16:10-16:40 Data-Driven Estimation
Speaker: Tom Oomen, Eindhoven University of Technology, The Netherlands
Description: Traditional ILC uses model-based designs to learn optimal command signals that minimize the error signal. Here, we will outline data-driven techniques to learn both command signal and other system properties, including system norms.
16:40-17:00 Ying Tan, moderator: Open Discussion and Concluding Remarks Discussion