Paperlane

no. 16 · Tuesday 17 November 2026

Reference Replay for Excitation Injection and Subsequent Exponential Convergence under Concurrent Learning Adaptive Control

Tom Kaufmann and Johann Reger

math.OC

Abstract

Concurrent learning (CL) guarantees precise identification of unknown parameters—if data with finite excitation (FE) happens to be collected online. To improve the reliability of CL, this work establishes a mechanism allowing FE-occurrence to be verified a priori: We propose replaying a given training segment in the reference until data with finite excitation is collected online. Provided the training segment fulfills a certain off-line excitation test, FE then occurs in unknown, but finite time. Moreover, we provide an upper bound on the time-to-FE. Even though this bound involves a dependency on the unknown initial Lyapunov-function value, its structure shows that FE occurs in arbitrarily early replay cycles when increasing the control gain; which may motivate employing a high-gain strategy. The theoretical findings are verified on an academic simulation example.

fields
Optimization and Control
pdf ↗ page arXiv:2611.21314 ↗