Safe Autonomy with Control Barrier Functions

Wei Xiao, Christos G. Cassandras, Calin BeltaSynthesis 2023 · DOI · OpenAlex

Abstract

As autonomous systems transition from laboratory environments to real-world applications like self-driving cars and human-centric manufacturing, the need for formal safety guarantees becomes paramount. This work explores a unified framework for safe autonomy centered on Control Barrier Functions (CBFs). By translating complex safety constraints into real-time Quadratic Programs (QPs), the methodology ensures that systems remain within defined safe states without sacrificing performance or computational efficiency. We examine technical advancements including High-Order CBFs for complex dynamics and machine learning techniques to enhance feasibility robustness, providing a robust bridge between high-level safety theory and practical embedded control.

Introduction

The deployment of autonomous systems in dynamic, human-populated environments necessitates a shift from traditional simulation-based testing to formal safety verification. The primary challenge lies in ensuring "forward invariance"—guaranteeing that a system starting in a safe state will never exit it—while maintaining the agility and speed required for practical utility. This paper synthesizes research into a framework that addresses the "triple threat" of control design: safety, performance, and computational complexity.

Methodology

The core of the proposed framework is the mapping of Control Barrier Functions to Quadratic Programs (CBF-to-QP). Safety requirements, typically expressed as state-variable inequalities, are transformed into linear constraints on control inputs. This allows the safety problem to be solved as a mathematically efficient QP, suitable for high-frequency execution on embedded hardware.

Technical Advancements

  • High-Order CBFs (HOCBFs): Standard CBFs often struggle with systems where the control input doesn't directly affect the safety constraint (e.g., controlling acceleration to maintain a position-based distance). HOCBFs generalize the framework to handle these high-relative-degree constraints.
  • Feasibility Robustness via Machine Learning: A common failure in safety-critical control is "infeasibility"—where no possible control action can satisfy the safety constraint. Machine learning is used to parameterize CBFs and learn values that maximize the "feasibility robustness" of the system.
  • Adaptive and Event-Driven CBFs: To handle noisy environments and unknown system dynamics, the framework introduces adaptive mechanisms and event-triggered execution, which updates control actions only when necessary rather than at fixed time intervals.

Key Insights

The fundamental principle of CBFs is that safety is guaranteed if the system's state trajectory never points outward at the boundary of the safe set. This approach minimizes the "safety tax"—the performance loss associated with conservative control—compared to traditional reachability analysis. By treating feasibility as a primary design objective, the framework acknowledges real-world physical constraints, such as maximum braking force, which are often overlooked in purely theoretical models.

Results and Evaluation

The framework has been validated across various domains, including simulated traffic scenarios (e.g., roundabout coordination and signal-free intersections for Connected Autonomous Vehicles) and robotics benchmarks (e.g., quadrotor navigation).

Key outcomes include:

  • Real-Time Tractability: The conversion of safety constraints into QPs allows for high-frequency control loops necessary for agile robotics and high-speed driving.
  • Provable Safety in CAVs: The framework effectively manages complex coordination in "internet of vehicles" scenarios, potentially addressing the vast majority of accidents currently attributed to human error.
  • Safety Shielding: The "Optimal Control with CBF" (OCBF) approach successfully tracks ideal, pre-planned trajectories while providing a safety intervention that triggers only when constraints are threatened.

Limitations and Future Work

While the CBF framework provides a powerful tool for safe autonomy, it remains dependent on reasonably accurate system models. Future research must address "infeasibility traps" where sudden environmental changes (e.g., an obstacle appearing within braking distance) make safety physically impossible, requiring "graceful degradation" or emergency maneuver strategies. Additionally, bridging the gap between continuous-time theory and discrete-time implementation remains a critical area for ensuring safety in ultra-high-precision applications.

Summary updated

BibTeX
@inproceedings{xiao2023safe,
  author = {Wei Xiao and Christos G. Cassandras and Calin Belta},
  title = {{Safe Autonomy with Control Barrier Functions}},
  booktitle = {Synthesis},
  year = {2023},
  doi = {10.1007/978-3-031-27576-0},
}
Figure