Barrier States Theory for Safety-Critical Multi-Objective Control

Hassan Almubarak, Nader Sadegh, Evangelos A. Theodorou

Abstract

Ensuring safety in autonomous systems often involves complex constrained optimization. This paper introduces Barrier States (BaS), a framework that treats safety as a dynamic state variable rather than an external constraint. By augmenting system dynamics with these safety margins, safety-critical control is transformed into a standard unconstrained stabilization problem, enabling the use of classical control tools like PID and LQR with formal safety guarantees.

Introduction

In safety-critical applications such as autonomous driving or industrial robotics, controllers must simultaneously achieve performance goals (e.g., reaching a target) and safety requirements (e.g., obstacle avoidance). Traditional approaches often rely on Control Barrier Functions (CBFs) within Quadratic Programming (QP) safety filters. However, these methods frequently struggle with high relative degree systems, exhibit myopic behavior that can lead to deadlocks, and are often incompatible with legacy control architectures like PID.

Methodology: The Barrier State Framework

The core innovation presented is the Safety Embedded System. Instead of checking constraints at every step, the authors define a Barrier State (BaS), denoted as $z$, derived from a barrier function (e.g., inverse or logarithmic penalties). This state is designed to grow unbounded as the system approaches the boundary of the safe set.

By appending the dynamics of $z$ to the original system, the authors create an augmented, unconstrained system. The control objective is then reformulated: stabilizing the augmented system to the origin ensures that $z$ remains bounded, which mathematically guarantees that the system never exits the safe set. This approach is "relative degree agnostic," making it applicable to complex mechanical systems where inputs do not directly affect safety metrics. Furthermore, it allows for the integration of safety directly into standard control loops, leading to the development of the Proportional-Integral-Derivative-Barrier (PIDB) controller.

Experimental Results

The BaS framework was evaluated through high-fidelity simulations, including Adaptive Cruise Control (ACC) and multi-robot collision avoidance.

One primary test case involved point-mass robot navigation around a circular obstacle. As shown in Figure 1, the BaS controller successfully guides the robot from various starting positions to a goal while maintaining a safe distance from the obstacle. The trajectories smoothly deform as they approach the danger zone, demonstrating how the barrier state naturally prioritizes safety over performance when necessary.

Trajectories of a point-mass robot navigating around a circular obstacle using the Barrier State framework.

In ACC simulations, the PIDB controller maintained a safe time-headway even under erratic lead vehicle behavior and significant sensor noise. The framework also proved effective in enforcing input constraints, such as maximum braking limits.

Discussion and Conclusion

The Barrier States theory offers a "safety-as-physics" perspective, where safety is baked into the system's own dynamics. By making the barrier state a part of the state vector, the controller perceives approaching a boundary as a physical deviation that must be corrected. This "repulsive" force naturally scales: when far from danger, the barrier state is negligible, allowing the performance controller to dominate. As the boundary is approached, the barrier state grows exponentially, forcing the controller to prioritize safety.

A significant advantage of this framework is its "legacy-friendly" nature. Because the augmented system is mathematically unconstrained, engineers can apply well-established linear control techniques like pole placement or LQR. This bridges the gap between advanced safety theory and industrial practice, where PID controllers remain the standard.

While the approach introduces additional state variables—one for each constraint—and inherent nonlinearities in the barrier dynamics, it provides a robust path for upgrading autonomous systems with formal safety guarantees. Future work will focus on managing the increased dimensionality in systems with numerous constraints and further exploring the theoretical links between BaS and Control Lyapunov-Barrier Functions (CLBFs).

Summary updated

BibTeX
@article{almubarak2025barrier,
  author = {Hassan Almubarak and Nader Sadegh and Evangelos A. Theodorou},
  title = {{Barrier States Theory for Safety-Critical Multi-Objective Control}},
  journal = {IEEE Trans. Automatic Control},
  year = {2025},
  doi = {10.1109/tac.2025.3584908},
}
Figure