Recurrent Control Barrier Functions: A Path Towards Nonparametric Safety Verification

Jixian Liu, Enrique Mallada

Abstract

Ensuring the safety of complex dynamical systems often relies on Hamilton-Jacobi (HJ) Reachability Analysis or Control Barrier Functions (CBFs). Both methods require computing a function that characterizes a safe set that can be made (control) invariant. However, the computational burden of solving high-dimensional partial differential equations (for HJ Reachability) or large-scale semidefinite programs (for CBFs) makes finding such functions challenging. In this paper, we introduce the notion of Recurrent Control Barrier Functions (RCBFs), a novel class of CBFs that leverages a recurrent property of the trajectories, i.e., coming back to a safe set, for safety verification. Under mild assumptions, we show that the RCBF condition holds for the signed-distance function, turning function design into set identification. Notably, the resulting set need not be invariant to certify safety. We further propose a data-driven nonparametric method to compute safe sets that is massively parallelizable and trades off conservativeness against computational cost.

Introduction

Autonomous systems, ranging from robotic manipulators to self-driving vehicles, require rigorous safety guarantees to operate in real-world environments. Traditionally, safety is framed as finding a "control invariant" set—a region of the state space from which the system is guaranteed never to exit. Two primary frameworks dominate this field: Hamilton-Jacobi (HJ) Reachability Analysis and Control Barrier Functions (CBFs).

Despite their theoretical elegance, both approaches face significant scalability issues. HJ reachability suffers from the curse of dimensionality, as it requires solving complex partial differential equations (PDEs) on a grid. CBFs, while more flexible, often require expensive Sum-of-Squares (SOS) optimization to synthesize valid barrier functions. This paper introduces Recurrent Control Barrier Functions (RCBFs) to address these challenges by relaxing the strict requirement of invariance to a more flexible property of recurrence.

Methodology: From Invariance to Recurrence

The core innovation of this work is the shift from invariance (never leaving a set) to recurrence (returning to a set within a finite time $\tau$). This relaxation allows the system to temporarily leave a designated safe set, provided it is guaranteed to return before encountering an unsafe region.

Conceptual illustration of a recurrent set and trajectory. In (a), a single set is recurrent if trajectories return to it. In (b), a collection of sets can be recurrent if trajectories cycle between them.

Signed-Distance as RCBFs

A major theoretical contribution of this paper is proving that a simple signed-distance function to a recurrent set can serve as a valid RCBF under mild conditions. This discovery transforms the difficult problem of synthesizing complex barrier functions into a more straightforward set identification problem.

Robust Cell Verification

To implement this theory, we developed a nonparametric, sampling-based algorithm. By utilizing Lipschitz continuity bounds on system dynamics, we can verify the safety of entire continuous regions (cells) based on a finite number of sampled trajectories. This bridges the gap between discrete data and continuous safety guarantees. The algorithm adaptively refines the grid, splitting ambiguous cells to increase resolution where needed. Because each cell is evaluated independently, the process is massively parallelizable on modern GPUs.

Experimental Results

We evaluated the RCBF framework on a 3D aircraft evasion problem, comparing it against a state-of-the-art JAX-based HJ reachability solver. The state space involves the relative planar location and orientation between two aircraft.

Comparison between HJ Reachability (a) and the proposed Recurrent Set Approximation (b) across different grid resolutions ($r_{min}$).

Safety and Efficiency

Our results demonstrate two primary advantages of the RCBF approach:

  1. Robustness at Low Resolution: At a coarse resolution ($r_{min}=1.111$), HJ reachability failed significantly, capturing only 51% of the unsafe zone and potentially misclassifying dangerous states as safe. In contrast, the RCBF method successfully captured 100% of the unsafe zone by providing a safe over-approximation.
  2. Computational Speed: At high resolutions ($r_{min}=0.041$), the RCBF method achieved a significant speedup, computing the safe set in 19.75 seconds compared to 83.73 seconds for the HJ PDE solver.

Furthermore, we observed a tunable trade-off: decreasing the return time parameter $\tau$ results in a tighter, less conservative safe set but increases the computational load.

Conclusion

Recurrent Control Barrier Functions offer a promising new direction for safety verification. By replacing the strict requirement of invariance with recurrence and leveraging signed-distance functions, we enable a nonparametric approach that is both faster and safer at low resolutions than traditional methods. Future work will focus on reducing the inherent conservativeness of the method and developing adaptive sampling strategies to further optimize computational performance.

Summary updated

Abstract

Ensuring the safety of complex dynamical systems often relies on Hamilton-Jacobi (HJ) Reachability Analysis or Control Barrier Functions (CBFs). Both methods require computing a function that characterizes a safe set that can be made (control) invariant. However, the computational burden of solving high-dimensional partial differential equations (for HJ Reachability) or large-scale semidefinite programs (for CBFs) makes finding such functions challenging. In this paper, we introduce the notion of Recurrent Control Barrier Functions (RCBFs), a novel class of CBFs that leverages a recurrent property of the trajectories, i.e., coming back to a safe set, for safety verification. Under mild assumptions, we show that the RCBF condition holds for the signed-distance function, turning function design into set identification. Notably, the resulting set need not be invariant to certify safety. We further propose a data-driven nonparametric method to compute safe sets that is massively parallelizable and trades off conservativeness against computational cost.
BibTeX
@misc{liu2025recurrent,
  author = {Jixian Liu and Enrique Mallada},
  title = {{Recurrent Control Barrier Functions: A Path Towards Nonparametric Safety Verification}},
  year = {2025},
  eprint = {2510.02127},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2510.02127},
}
Figure