Meta-Learning for Data-Driven Control System Design: Theory and Applications
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This Brief shows how similarity across dynamical systems can accelerate data-driven controller and estimator design. Classical data-driven control (DDC) methods avoid explicit modeling but remain plant-specific and require costly re-tuning whenever operating conditions change. We propose two complementary strategies that exploit prior designs and perturbed training classes. First, Meta-DDC reuses controllers tuned on similar systems, and its extension Meta-AutoDDC automates the selection of the reference model. Together they guarantee non-deteriorating performance and improved robustness when designing controllers for unseen but related plants [ 4 , 7 ]. Second, in-context learning with transformer architectures produces contextual controllers and estimators, trained on perturbed simulations to generalize across entire classes of nonlinear systems without re-training [ 5 ]. Experiments on brushless motors and nonlinear process benchmarks confirm faster design, reduced data requirements, and improved robustness. We conclude with perspectives on scalable, similarity-aware control design.
Fields of Study & Subjects
Control Systems and Identification
Iterative Learning Control Systems
Advanced Control Systems Optimization
Publication details
- Journal / source
- SpringerBriefs in applied sciences and technology
- Pages
- 65–74
- Publication date
- 26 March 2026
- Language
- EN
- Document type
- Book chapter