Reinforcement Learning-Based Safety-Critical Control of Input Affine Nonlinear Systems
IEEE Transactions on Neural Networks and Learning Systems · 2026 · 1-12
DOI: 10.1109/tnnls.2026.3748522 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Neural Networks and Learning Systems, 2026. Verfasst von Fan Geng und 3 weiteren Autor:innen. Gefördert von: National Natural Science Foundation of China.
Abstract (Originalsprache)
This article is specialized in handling the safety-critical control problem of the nonlinear system, which extends existing control barrier function-quadratic programming (CBF-QP) or control Lyapunov barrier function (CLBF)-based safety control methods into scenarios with unknown drift dynamics by enhancing the system's online learning ability. The technical challenge mainly lies in developing a model-free safety-critical control approach with the aid of control barrier function (CBF), reinforcement learning techniques, and output regulation theory. To be specific, we first propose a bounded reinforcement learning-based tracking controller (BRLTC), composed of an estimated steady-state controller and an approximate optimal feedback controller. The former is based on the neural network identifier and projection-based adaptive laws for compensating unknown dynamics, while the latter is synthesized by optimizing a performance index for stabilization. Then, a safeguarding controller is constructed based on a Lyapunov-like CBF in a decoupled manner. Our design prioritizes system safety, and trajectory tracking with bounded error is achieved if there is no conflict in the multiple control objectives.
Förderung
- National Natural Science Foundation of China
Zitieren
APA
Geng, F., Dong, H., Dong, Y., & Xu, S. (2026). Reinforcement Learning-Based Safety-Critical Control of Input Affine Nonlinear Systems. IEEE Transactions on Neural Networks and Learning Systems, 1-12. https://doi.org/10.1109/tnnls.2026.3748522
Vancouver
Geng F, Dong H, Dong Y, Xu S. Reinforcement Learning-Based Safety-Critical Control of Input Affine Nonlinear Systems. IEEE Transactions on Neural Networks and Learning Systems. 2026:1-12. doi:10.1109/tnnls.2026.3748522
BibTeX
@article{Geng2026,
author = {Geng, Fan and Dong, Hairong and Dong, Yi and Xu, Shengyuan},
title = {Reinforcement Learning-Based Safety-Critical Control of Input Affine Nonlinear Systems},
journal = {IEEE Transactions on Neural Networks and Learning Systems},
year = {2026},
pages = {1--12},
doi = {10.1109/tnnls.2026.3748522}
}
EUVIMED