Regions of Attraction of CBF-Based Safety Filters for Stable Planar Linear Systems

summary

The gist

The gist In this paper we study the dynamical properties induced by safety filters based on control barrier functions (CBFs) for stable planar linear systems.

In short

This study investigates how safety filters, based on control barrier functions (CBFs), affect the dynamics of a simple planar linear system with an affine constraint. It shows that even in this basic setting, the filtered system can oscillate or diverge while remaining safe. The research fully characterizes conditions for global stability and identifies how the filter creates saddle points on constraint boundaries.

Key concepts

Control Barrier Functions (CBFs): CBFs are mathematical functions used to define control inputs that guarantee a system stays within a desired safe region. They are characterized by the condition h˙(x, u) + αh(x) ≥ 0. The safety filter uses these functions to select an input that keeps the system safe.
Safety Filter: A safety filter minimally modifies a nominal controller only when necessary to ensure the system respects hard constraints defined by CBFs. It works by selecting an admissible input from a set of inputs closest to the nominal command, leaving it unchanged if it already satisfies safety conditions.
Planar Linear System with Affine Constraint: This is the simplified model used for analysis, involving a system described by linear dynamics (x˙ = Ax + Bu) in two dimensions (planar). The constraint is an affine condition of the form h(x) = c⊤x + d. This setup allows researchers to rigorously study how safety filters impact stability.
Region of Attraction: This refers to the set of initial states from which a system will remain bounded or converge to a desired equilibrium. The paper characterizes this region for the filtered system, showing that depending on parameters like damping ratio and filter gain, this region can be globally stable, asymptotic stable, or limited by unstable orbits.

Terminology used across episodes

Reinforcement Learning (RL) — explained in 25 episodes
Sample Complexity — explained in 20 episodes
Retrieval-Augmented Generation (RAG) — explained in 19 episodes
Federated Learning — explained in 17 episodes
Flow Matching — explained in 17 episodes
Quantum Fisher Information (QFI) — explained in 16 episodes

This episode discusses

The paper

Regions of Attraction of CBF-Based Safety Filters for Stable Planar Linear Systems · Read on arXiv

Adam K. Kiss, Pol Mestres, Andras Bartfai, Shima Sadat Mousavi, Aaron D. Ames

California Institute of Technology · Budapest University of Technology and Economics

In this paper we study the dynamical properties induced by safety filters based on control barrier functions (CBFs). These are obtained by minimally modifying a nominal controller with the goal of rendering it safe. We consider the simple setting of a planar linear plant with a stable linear nominal controller and an affine constraint. We show that even in this simple setting, the filtered system can oscillate or diverge while remaining safe. We fully characterize the parameters for which the origin is globally (exponentially or asymptotically) stable. For all other parameters, the filter creates a saddle on the constraint boundary, and we characterize the region of attraction of the origin. For underdamped nominal dynamics, its boundary is an unstable periodic orbit, a homoclinic loop, or the saddle's stable manifold, depending on whether the CBF gain is below, at, or above a homoclinic transition value. A vehicle sideslip example and two three-dimensional examples illustrate the results and what changes beyond planar dynamics.

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: Today's paper: "Regions of Attraction of CBF-Based Safety Filters for Stable Planar Linear Systems".

Dev: The gist In this paper we study the dynamical properties induced by safety filters based on control barrier functions (CBFs) for stable planar linear systems.

Rosa: First, who's behind it and why it matters.

Title and authors: Rosa: So we're looking at this paper called "Regions of Attraction of CBF-Based Safety Filters for Stable Planar Linear Systems Adam K. Kiss, Pol Mestres, Andras Bartfai, Shima Sadat Mousavi, and Aaron D. Ames" and what it means for us in the real world.

Dev: It's about how we use control barrier functions or CBFs to make a nominal controller safe by only changing the input when absolutely necessary.

Taro: So, basically, they take a system that's supposed to be stable and add this safety filter layer on top of it.

Rosa: And what they show is that even in this simple setting—a planar linear plant with an affine constraint—the filtered system can actually start oscillating or diverge while still being safe.

Dev: That's a key point because the nominal closed loop dynamics and the filtered closed loop dynamics can be quite different, which means we have to look at both.

Taro: So, this paper is really pushing us to think about what happens when we introduce that switching between the original controller and this corrective filter control.

Rosa: And it sets up a framework to fully characterize exactly what parameters make the origin globally stable in these scenarios.

The paper's summary: Dev: They reduce this whole planar safety-filtered closed loop down to a dimensionless normal form with three specific parameters: the damping ratio ζ of the stable nominal dynamics, a normalized invariant-zero parameter k, and the filter gain α.

Taro: And that sign on k is really important because it dictates whether the active mode of the system is stable or not.

Rosa: If k is greater than zero, they prove global exponential stability for every possible filter gain alpha. That's a strong result building on previous work in this area seven <ref:2610.21221#pg1>.

Dev: But when k equals zero, they establish global asymptotic stability, but that's not quite exponential stability yet.

Taro: And if k is less than zero, the filter actually creates a saddle equilibrium right there on the constraint boundary.

Rosa: That saddle equilibrium opens up a whole new way to characterize the region of attraction for the origin in this case.

Dev: For underdamped nominal dynamics, they say that this boundary depends entirely on whether you have a low gain, an intermediate gain, or a high gain relative to a specific homoclinic transition value.

The paper's improvements: Taro: So the paper suggests some ways to make this analysis more concrete and useful for real systems.

Rosa: They point out that they can precisely characterize when the filter creates an undesired equilibrium or a limit cycle by identifying that specific "homoclinic transition value," which they say is given by an explicit scalar equation.

Dev: That’s helpful because it moves us from just saying there's a transition to giving us a formula we can actually use to find where that happens.

Taro: They also discuss how to determine if the region of attraction is bounded or unbounded for systems with nonminimum phase constraints by checking if the nominal loop carries Qk to Qalpha, using this condition in equation (eighteen).

Rosa: That helps us predict the long-term behavior before we even start running complex simulations.

Dev: And they also suggest an idea for real-time safety assessment and predictive control where you can use these derived conditions to predict if a trajectory will enter an unsafe region based on whether the filter is active or inactive.

Conclusion: Rosa: So to wrap up, this paper on "Regions of Attraction of CBF-Based Safety Filters for Stable Planar Linear Systems" shows that safety filters can really alter the global behavior of a stable planar loop by creating new equilibria and oscillations.

Dev: The core finding is that we have three distinct regimes based on the parameter k: global exponential stability when k is positive, asymptotic stability when k is zero, and a saddle equilibrium when k is negative.

Taro: And for the underdamped case, the boundary of attraction can be an unstable periodic orbit or a homoclinic loop depending on where your filter gain sits relative to that transition value.

Rosa: This gives us specific things to look for in our designs—oscillations or saddle points—and it shows that recovery from very large safe states is only possible if the nominal loop doesn't oscillate and the filter gain is at least the decay rate of its slow mode.

Dev: We still have limitations here; they note that they can't fully classify things in three dimensions, and they also mention that for a vehicle sideslip problem, their results are just examples.

Taro: So we have a good starting point for planar analysis, but we're still waiting on the full three-dimensional classification to really understand everything about what happens when the world misbehaves.

More episodes

← Home