Authors
Mahmood Mazare, Majid Mokhtari, Mostafa Taghizadeh, Hossein Ramezani
Published in
ISA transactions. Nov 11, 2026. Epub Nov 11, 2026.
Abstract
Enforcing safety for nonlinear systems that combine high-relative-degree state constraints, hard actuator saturation and unmodeled dynamics remains an open problem: reactive safety filters presume unbounded control authority, while robust designs based on static worst-case bounds are unnecessarily conservative. This paper proposes the Augmented Gaussian Process Koopman Control Barrier Function (AGPK-CBF) framework. A system augmentation strategy promotes the physical input to a state, converting input constraints into barriers of relative degree one that are exactly disturbance-free and rendering the auxiliary jerk input available for optimization. Koopman operator theory lifts the augmented dynamics into a linear predictor, enabling model predictive control to generate smooth nominal trajectories, while Gaussian Process regression supplies a posterior mean for feed-forward compensation and a posterior variance for probabilistic error bounding. A continuous adaptation law scales the robustness margin with proximity to the safety boundary, inflating conservatism only where epistemic uncertainty warrants it. We establish probabilistic forward invariance, deterministic actuator compliance, Input-to-State Stability and recursive feasibility under time-varying margins. Benchmarks against three state-of-the-art designs and a component-wise ablation yield a 0.00% constraint violation rate, strict torque-limit compliance and chatter-free actuation.
PMID:
43052125
Bibliographic data and abstract were imported from PubMed on 19 Nov 2026.
Read full publication at:
Please sign in
to see all details.
Advertisement
Stats
- Recommendations n/a n/a positive of 0 vote(s)
- Views 3
- Comments 0