Designing Control Barrier Functions Using a Dynamic Backup Policy

Basic Information

• Paper ID: 2510.09810
• Title: Designing Control Barrier Functions Using a Dynamic Backup Policy
• Authors: Victor Freire, Marco M. Nicotra (University of Colorado, Boulder)
• Classification: eess.SY cs.SY
• Publication Date: arXiv preprint, October 2025
• Paper Link: https://arxiv.org/abs/2510.09810

Abstract

This paper proposes a systematic method for constructing control barrier functions (CBFs) for nonlinear control-affine systems under arbitrary state and input constraints. Inspired by the reference governor literature, the method defines a family of backup policies parameterized by the system's equilibrium manifold. The control barrier function is defined on an augmented state-reference space: given a state-reference pair, the method quantifies the distance to constraint violation at any future time while maintaining the backup policy reference constant. Sensitivity analysis is then employed to compute the (possibly non-smooth) Jacobian matrix with respect to the augmented state vector.

Research Background and Motivation

Core Problem

Control barrier functions (CBFs) are powerful tools for designing safety-critical system controllers, but systematic synthesis for general systems under arbitrary state and input constraints remains challenging. Although the modern definition of CBFs was established in 2017, their systematic design remains an open problem.

Problem Significance

In many applications, practitioners must resort to candidate CBFs because they are easier to design and can achieve good performance through parameter tuning and slack variables, but lack rigorous safety guarantees of effective CBFs.

Limitations of Existing Approaches

1. Machine Learning Methods: Lack theoretical guarantees
2. Sum-of-Squares Methods: Applicable only to polynomial systems
3. Expert Demonstration Methods: Dependent on high-quality training data
4. Backup CBF Methods: Lack systematic design methodology for backup policies

Research Motivation

The authors' prior work demonstrated that dynamic safety margins (DSMs) in augmented state-reference space are CBFs. This paper extends trajectory-based DSMs to construct CBFs, providing a more systematic and general approach.

Core Contributions

1. Systematic CBF Design Method: Proposes a trajectory prediction-based DSM-CBF construction method applicable to nonlinear control-affine systems with arbitrary state and input constraints
2. Dynamic Backup Policy: Backup policies are parameterized by the system's equilibrium manifold with time-varying parameterization, providing greater flexibility than fixed backup policies
3. Theoretical Guarantees: Proves asymptotic stability of the underlying sensitivity matrix under mild assumptions and rigorously addresses issues arising from potential non-smoothness of CBFs
4. Practical Considerations: Addresses finite-time horizon and smoothness issues, proposing more implementable strategies with proven sufficiency and feasibility

Methodology Details

Problem Formulation

Consider a control-affine system:

ẋ = f(x) + g(x)u

subject to state constraints x ∈ X and input constraints u ∈ U, with the objective of designing a CBF to enforce constraint satisfaction.

Core Assumptions

Assumption 3: The system admits an equilibrium manifold parameterized by v ∈ ℝˡ, with continuous functions x̄(v) and ū(v) such that:

f(x̄(v)) + g(x̄(v))ū(v) = 0, ∀v ∈ ℝˡ

and there exists a pre-stabilizing control policy π(x,v) making x̄(v) an asymptotically stable equilibrium point.

Trajectory-Based DSM-CBF Construction

1. Dynamic Safety Margin Definition

Δ(x,v) = inf_{τ∈[0,∞)} c(Φ(τ,x,v), v)

where:

• c(x,v) describes the reference-dependent state constraint
• Φ(τ,x,v) is the solution of the pre-stabilizing dynamics

2. Sensitivity Analysis

Compute the flow sensitivity Jacobian matrices:

State Sensitivity:

∂S_x(τ)/∂τ = (∂f_π(Φ(τ),v)/∂x)S_x(τ), S_x(0) = I_n

Reference Sensitivity:

∂S_v(τ)/∂τ = (∂f_π(Φ(τ),v)/∂x)S_v(τ) + ∂f_π(Φ(τ),v)/∂v, S_v(0) = 0

3. Finite-Time Horizon Implementation

Since infinite-time horizon is numerically infeasible, a finite horizon T is employed with terminal DSM:

Δ(x,v) = [min_{τ∈[0,T]} c(Φ(τ),v); Δ_T(Φ(T),v)]

Technical Innovations

1. Parameterized Backup Policy Family: Unlike fixed backup policies, parameterization via equilibrium manifold provides an entire family of policies
2. Augmented Space CBF: CBF defined in state-reference augmented space naturally handles reference dynamics
3. Non-Smoothness Handling: Rigorously addresses potential non-smoothness of CBF using Clarke generalized Jacobian matrices
4. Feasibility Guarantee: Proves feasibility of the safety filter, where u = π(x,v), w = 0 is always a feasible solution

Experimental Setup

System Model

Inverted pendulum on cart system:

M(q)q̈ + C(q,q̇)q̇ + G(q) = Bu

Parameters:

• Cart mass m_c = 1 kg
• Pendulum mass m_p = 0.5 kg
• Pendulum length L = 0.7 m

Constraints

• State constraints: |x| ≤ 4.5 m, |θ-π| ≤ π/9
• Input constraints: |u| ≤ 20 N

Controller Design

• Pre-stabilizing controller: LQR design, K_π = -0.44, 35.3, -1.4, 8.0
• Nominal controller: K_κ = -35, 150, -20, 50
• Prediction horizon: T = 10 seconds
• Class K∞ functions: α(c) = 100c (constraints), α(c) = 400c (terminal DSM)

Comparison Methods

1. Nominal Control: Using only nominal controller (unsafe)
2. Lyapunov-Based DSM-CBF: DSM method based on Lyapunov functions
3. ERG: Explicit Reference Governor
4. Backup CBF: Traditional backup CBF method

Experimental Results

Main Results

Simulation results demonstrate:

1. Safety: The proposed method successfully maintains all constraint satisfaction, with cart position remaining within ±4.5 m and pendulum angle deviation within ±20°
2. Performance Advantages:
   • Compared to Lyapunov-based DSM-CBF: Faster convergence and smoother trajectories
   • Compared to ERG: Significantly improved response speed
   • Compared to traditional backup CBF: Superior performance due to additional degrees of freedom from parameterized backup policy family
3. Control Input: Input always satisfies ±20 N constraints and exhibits smoother behavior compared to other methods

Performance Analysis

• Trajectory Quality: The proposed method achieves smooth transitions from initial position to target position
• Constraint Satisfaction: Strictly satisfies all state and input constraints
• Convergence: Exhibits better convergence characteristics compared to baseline methods

Theoretical Prediction Verification

Experiments verify the following theoretical expectations:

1. Parameterized backup policy family provides better performance than fixed backup policies
2. Trajectory prediction method is effective for systems with high relative degree and non-minimum phase characteristics
3. Practicality of augmented space CBF method

Related Work

CBF Design Methods

1. Machine Learning Approaches (Harms et al. 2024): Using machine learning to design CBFs
2. Sum-of-Squares Methods (Dai et al. 2024): Designing polynomial CBFs
3. Expert Demonstration Methods (Lindemann et al. 2024): Learning CBFs from safe demonstrations

Backup CBF Related Work

1. Chen et al. (2021): Introduced the concept of backup CBFs
2. Van Wijk et al. (2024): Disturbance-robust backup CBFs
3. Choi et al. (2021): Robust control barrier value functions

Reference Governor Literature

This paper's methodology is inspired by reference governor literature, particularly the dynamic safety margin concept, but extends it to the CBF framework.

Conclusions and Discussion

Main Conclusions

1. Systematic Design: Provides a systematic method for constructing effective CBFs applicable to nonlinear systems with arbitrary constraints
2. Theoretical Guarantees: Provides rigorous theoretical analysis and feasibility guarantees under mild assumptions
3. Practicality: The method is simple to implement with relatively low computational complexity

Limitations

1. Assumption Dependency: Requires the system to admit an equilibrium manifold and ability to design a pre-stabilizing controller
2. Computational Complexity: Although lower than MPC, still requires trajectory prediction and sensitivity computation
3. Parameter Tuning: Requires appropriate selection of class K∞ functions and prediction horizon

Future Directions

1. Robustness Analysis: Study robustness properties under bounded disturbances
2. Alternative Prediction Techniques: Explore other trajectory prediction techniques
3. Computational Optimization: Methods to further reduce computational complexity

In-Depth Evaluation

Strengths

1. Theoretical Rigor: Provides comprehensive theoretical analysis including existence, feasibility, and convergence proofs
2. Method Generality: Applicable to a broad class of nonlinear control-affine systems
3. Implementation Simplicity: Relatively lighter computational burden compared to MPC-based methods
4. Complexity Handling: Capable of handling input saturation, multiple constraints, and strong nonlinearity

Weaknesses

1. Assumption Limitations: The equilibrium manifold assumption may restrict the method's applicability
2. Parameter Sensitivity: Method performance may be sensitive to parameter selection
3. Limited Experimental Scope: Validation on only one system, lacking broader experimental verification

Impact

1. Academic Contribution: Provides new theoretical framework and practical methods for CBF design
2. Practical Value: Relatively simple method with potential for real-world system applications
3. Reproducibility: Authors provide open-source code, enhancing reproducibility

Applicable Scenarios

1. Safety-Critical Systems: Such as robotics and autonomous vehicles requiring strict constraint satisfaction
2. Nonlinear Systems: Particularly suited for nonlinear control systems with complex constraints
3. Real-Time Control: Lower computational complexity than MPC, suitable for real-time applications

References

The paper cites major works in the field, including:

• Ames et al. (2017): Modern definition of CBFs
• Chen et al. (2021): Concept of backup CBFs
• Nicotra and Garone (2018): Explicit reference governors
• Authors' prior work: Theoretical foundation of DSM-CBF

Overall Assessment: This is a high-quality paper with rigorous theory and novel methodology, providing a systematic approach to CBF design. While maintaining theoretical rigor, it demonstrates good practical applicability. The method's generality and relatively simple implementation give it significant practical value.