New model-free CBF framework enhances safety without system models

Control Barrier Functions (CBFs) are widely used to enforce safety constraints in nonlinear control systems, but they typically require accurate mathematical models of the system. That assumption breaks down when dynamics are unknown or heavily uncertain. A new arXiv paper (2608.15361) from researchers including Loïc Michel and Ricardo de Castro introduces a completely model-free framework for recursive CBFs. Instead of relying on an explicit model, the method uses the ultra-local model approach—an online estimation technique that approximates the system's unknown dynamics in real time. This lets the controller construct CBF constraints without prior knowledge of the plant.

The approach addresses systems with higher relative degree, where safety constraints involve derivatives of the output, and it enables both enforcement and anticipation of safety limits. The authors show that the method enhances robustness to disturbances and model mismatch, making it suitable for real-world applications where models are imperfect. They validate the framework on an adaptive cruise control benchmark, demonstrating that a vehicle can maintain safe following distances without using a predetermined dynamics model. This is a step toward more practical safety-critical control for autonomous systems, especially in scenarios where modeling is costly or impossible. The full paper is available on arXiv with code links and references.