.png){kind=icon}
.png){kind=icon}
Return
Data-driven certified control barrier functions
.png){kind=icon}
PRE
AI
DOI：10.1016/j.sysconle.2025.106325.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Control barrier functions and invariant sets are fundamental to constraint enforcement in safety-critical applications. To ensure safety, data-driven control methods based on nonlinear system identification often assume the existence of a barrier function computed from data. This paper presents a wide artificial neural network trained under constrained output evolution to synthesize barrier functions that characterize invariant sets for locally Lipschitz continuous nonlinear systems. We embed safety guarantees in the training, in contrast to probabilistic guarantees often given by scenario optimization or post-hoc verification in trial-and-error methods. Further, guaranteeing invariance for nonlinear systems generally requires solving a model-based nonlinear and often non-convex program. Conversely, we formulate our approach as a linear program for nonlinear systems. To demonstrate the versatility of our method, we approximate the invariant set of the Julia map, a chaotic system known for its complex behavior and fractal invariant sets.
Keywords:
Learning-based control
Safe data-driven control
Constrained nonlinear control
Journal
S
IF:
2.5
Papers:
154
Citations:
0