Rectified Control Barrier Functions for High-Order Safety Constraints

Pio Ong, Max H. Cohen, Tamás G. Molnár, Aaron D. AmesIEEE Control. Syst. Lett 2024 · DOI

TL;DR

This paper introduces Rectified Control Barrier Functions (ReCBFs), a novel method for synthesizing safety controllers for systems with high-order safety constraints. ReCBFs overcome limitations of existing High-Order Control Barrier Functions (HOCBFs) by activating safety enforcement only when necessary, leading to more robust and well-defined controllers, especially in challenging scenarios.

What Problem Does This Paper Address?

Ensuring the safety of dynamical systems is paramount, especially in critical applications like robotics and autonomous vehicles. Control Barrier Functions (CBFs) provide a powerful mathematical framework for guaranteeing that a system's state remains within a predefined safe set. However, constructing CBFs for safety constraints that depend on higher-order derivatives of the system's state (high relative degree constraints) or involve multiple inputs with different "orders of influence" is challenging. Existing methods like High-Order Control Barrier Functions (HOCBFs) can struggle with these complex constraints, sometimes leading to ill-defined controllers or limitations in stability and robustness. This paper aims to bridge this gap by proposing a new CBF construction that is more versatile and robust for these challenging safety requirements.

Methods

The core innovation of this paper is the introduction of Rectified Control Barrier Functions (ReCBFs). The key idea is to build upon the HOCBF framework but incorporate an "activation function" that intelligently decides when to enforce the safety constraint.

  • Rectification Mechanism: Instead of always enforcing a condition derived from higher-order derivatives, ReCBFs use a Rectified Linear Unit (ReLU) function. This means the safety enforcement mechanism is only "activated" when the system's state is approaching an unsafe region, as determined by an intermediate safety function.
  • Weak Relative Degree Handling: The paper specifically addresses constraints with "weak relative degree," where the input might not affect all derivatives of the safety function. ReCBFs are designed to handle these situations more gracefully than traditional HOCBFs.
  • Iterative Construction: For safety constraints with higher relative degrees (greater than two), the ReCBF construction is applied iteratively, building upon intermediate safety functions.
  • Theoretical Guarantees: The paper provides theoretical proofs demonstrating that ReCBFs are indeed valid CBFs, ensuring forward invariance of the safe set and inheriting desirable properties like stability and robustness.

Datasets

Not discussed in the paper. The paper focuses on theoretical development and validation through simulations and a specific aircraft control problem.

Key Insights and Ideas

  • "Lazy" Safety Enforcement: The most significant insight is the concept of "rectifying" safety conditions. By only activating the safety enforcement when truly needed (i.e., when the state is approaching unsafe territory), ReCBFs avoid unnecessary control actions and potential issues that arise from trying to enforce safety too aggressively.
  • Bridging HOCBFs and CBFs: ReCBFs offer a way to construct true CBFs from HOCBF-like conditions. This means that established CBF theory regarding stability and robustness can be more readily applied.
  • Handling Singularities: The paper demonstrates that ReCBFs can handle scenarios where traditional HOCBFs fail, particularly when the input's influence on the safety function becomes zero (a singularity). This is crucial for practical applications where such singularities are common.
  • Unified Framework: The proposed ReCBF framework is shown to be applicable to various scenarios, including those with relative degree two, mixed-input relative degrees, and higher relative degrees, offering a more unified approach.

Key Results

The paper presents theoretical results proving the efficacy of ReCBFs. Through illustrative examples and a fixed-wing aircraft control problem, the authors demonstrate:

  • Safety Guarantees: ReCBFs successfully enforce safety constraints, keeping the system within the desired safe set.
  • Improved Robustness: Compared to HOCBFs, ReCBFs lead to controllers that are well-defined even at points where the input's influence on the safety function is zero.
  • Enhanced Performance: In the aircraft example, the ReCBF-based controller tracks the desired trajectory more effectively and for a longer duration than the HOCBF-based controller, which experiences controller failure.
  • Stability: The paper suggests that ReCBFs can lead to asymptotically stable closed-loop systems, a property not always guaranteed by HOCBFs.

Limitations and Future Work

  • Hyperparameter Sensitivity: The authors acknowledge that ReCBF controllers can be sensitive to the choice of hyperparameters (e.g., the activation functions $\alpha$ and $\gamma$). Improper tuning can lead to controllers with large Lipschitz constants and consequently large control inputs.
  • Future Directions:
    • Hyperparameter Tuning: Further research is needed to characterize the impact of hyperparameters and develop systematic tuning strategies.
  • Unified Framework: Unifying the results for mixed and high relative degrees into a single, more general framework is a key future direction.
  • Practical Implementation: Exploring the application of ReCBFs to more complex, real-world systems and investigating their performance in the presence of noise and uncertainties would be valuable.

Summary updated

BibTeX
@article{ong2024rectified,
  author = {Pio Ong and Max H. Cohen and Tamás G. Molnár and Aaron D. Ames},
  title = {{Rectified Control Barrier Functions for High-Order Safety Constraints}},
  journal = {IEEE Control. Syst. Lett},
  year = {2024},
  doi = {10.1109/lcsys.2024.3518393},
}
Figure