arxiv: 2604.08831 · v1 · submitted 2026-04-10 · 📡 eess.SY · cs.SY

Recognition: 2 theorem links

· Lean Theorem

Probabilistic Control Barrier Functions for Systems with State Estimation Uncertainty using Sub-Gaussian Concentration

Kazuya Echigo , David E. J. van Wijk , Pol Mestres , Ersin Da\c{s} , Joel W. Burdick , Aaron D. Ames

Authors on Pith no claims yet

Pith reviewed 2026-05-10 18:14 UTC · model grok-4.3

classification 📡 eess.SY cs.SY

keywords probabilistic control barrier functionssub-Gaussian concentrationstate estimation uncertaintyparticle-based CVaRfinite-sample boundssafety-critical controlstochastic systems

0 comments

The pith

Gaussian uncertainties preserve sub-Gaussianity in barrier functions, enabling finite-sample safety certificates via particle CVaR estimates.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a particle-based probabilistic control barrier function method for systems whose state estimates contain Gaussian noise. It proves that such noise, when passed through Lipschitz-continuous control-affine dynamics, keeps the barrier-function increment sub-Gaussian and supplies explicit tail bounds. These bounds are then used to quantify the approximation error that arises when conditional value at risk is estimated from a finite number of particles. The resulting optimization problem therefore carries finite-sample safety certificates that are both tractable and less conservative than earlier stochastic barrier approaches.

Core claim

The authors establish that Gaussian uncertainties propagating through Lipschitz-continuous control-affine dynamics preserve sub-Gaussianity of the barrier function increment, with explicit tail bounds. Leveraging this structure, they derive finite-sample bounds on the approximation error between particle-based CVaR estimates and ground-truth probabilistic constraints; applying this yields a tractable optimization problem formulation with finite-sample safety certificates.

What carries the argument

Sub-Gaussian concentration bounds on the barrier-function increment (its tails decay at least as fast as a Gaussian), which supply the tail probabilities needed to bound particle-approximation error for the CVaR safety constraint.

If this is right

A convex optimization problem can be solved online to enforce probabilistic safety with a user-chosen failure probability.
The number of particles required for a given error tolerance is finite and explicitly computable from the sub-Gaussian parameter.
The same concentration argument supplies certificates that are tighter than those obtained from moment-based or worst-case stochastic CBF formulations.
The framework applies to any control-affine plant whose state estimator produces Gaussian-distributed errors.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same tail-bound technique could be applied to other risk measures or to barrier functions defined on different state quantities.
If real sensor noise is only approximately sub-Gaussian, the derived particle count would still serve as a conservative starting value for validation experiments.
Combining the particle CVaR estimate with online learning of the dynamics Lipschitz constant would produce an adaptive version of the safety certificate.

Load-bearing premise

State-estimation uncertainties are exactly Gaussian and the dynamics are Lipschitz continuous and control-affine; without these the sub-Gaussian preservation and resulting error bounds no longer hold.

What would settle it

Run Monte Carlo trials on a control-affine system driven by non-Gaussian (e.g., Student-t) noise and check whether the observed tail decay of the barrier-function increment matches the explicit sub-Gaussian bound predicted by the paper.

Figures

Figures reproduced from arXiv: 2604.08831 by Aaron D. Ames, David E. J. van Wijk, Ersin Da\c{s}, Joel W. Burdick, Kazuya Echigo, Pol Mestres.

**Figure 2.** Figure 2: Performance metrics across Monte Carlo trials. (a) Barrier function [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

read the original abstract

Safety-critical control systems, such as spacecraft performing proximity operations, must provide formal safety guarantees despite stochastic uncertainties from state estimation and unmodeled dynamics. Although Control Barrier Functions (CBFs) have been extended to stochastic systems, existing approaches typically face a trade-off between the tightness of probabilistic guarantees and computational tractability. This paper presents a particle-based probabilistic CBF framework that overcomes this limitation by exploiting the sub-Gaussian structure of the barrier function increment under Gaussian uncertainties. We establish that Gaussian uncertainties propagating through Lipschitz-continuous control-affine dynamics preserve sub-Gaussianity of the barrier function increment, with explicit tail bounds. Leveraging this structure, we derive finite-sample bounds on the approximation error between particle-based Conditional Value at Risk (CVaR) estimates and ground-truth probabilistic constraints; applying this yields a tractable optimization problem formulation with finite-sample safety certificates. We show through numerical experiments how the proposed approach provides tight yet provably valid probabilistic safety guarantees.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

This paper gives a particle CVaR approximation for probabilistic CBFs under Gaussian noise with claimed finite-sample bounds from sub-Gaussian tails, but the guarantees may not automatically survive optimization over the control input.

read the letter

The main takeaway is that this work develops a particle-based probabilistic CBF method with explicit finite-sample bounds derived from sub-Gaussian concentration under Gaussian state estimation noise. It claims to deliver tractable optimization while preserving safety certificates. What stands out as new is the preservation of sub-Gaussianity for the barrier increment when Gaussian noise passes through Lipschitz control-affine dynamics, along with the resulting tail bounds and the error analysis for the particle CVaR approximation. This combination lets them formulate an optimization that comes with finite-sample guarantees, which is a step beyond many existing stochastic CBF approaches that either lack tight bounds or become intractable. The paper does well in laying out the assumptions clearly and showing numerical results that illustrate the approach on example systems. The derivations appear grounded in standard concentration tools applied to the CBF setting. The potential issue is whether those finite-sample bounds remain valid after optimization over the control input. Sub-Gaussian tail bounds typically apply to a fixed random variable, so for a chosen u. If the optimization can pick a u where the particle estimate deviates more than the bound predicts, the certificate could fail without an additional argument like a union bound over a discretization of the control space or a Lipschitz property of the CVaR with respect to u. The abstract suggests they obtain certificates for the optimization problem, but this detail needs verification in the full proofs. The work assumes Gaussian uncertainties and Lipschitz continuous control-affine dynamics; if those do not hold, the sub-Gaussian property and bounds do not apply. That is a standard modeling choice but limits applicability. This paper targets specialists in stochastic safety-critical control, such as those in robotics or autonomous systems. Readers interested in extending CBFs to uncertain environments would find the technical development and experiments worthwhile. It has enough substance to warrant peer review, as the core claims can be checked against the derivations and experiments. I recommend putting it through review rather than desk rejecting it.

Referee Report

1 major / 3 minor

Summary. The manuscript proposes a particle-based probabilistic Control Barrier Function (CBF) framework for safety-critical systems with Gaussian state estimation uncertainties. It establishes that such uncertainties, propagating through Lipschitz-continuous control-affine dynamics, preserve sub-Gaussianity of the barrier function increment with explicit tail bounds. From this, finite-sample bounds are derived on the approximation error between particle-based CVaR estimates and the ground-truth probabilistic constraints, yielding a tractable optimization problem with finite-sample safety certificates. The approach is illustrated via numerical experiments on systems such as spacecraft proximity operations.

Significance. If the uniformity of the finite-sample bounds over control inputs holds, the work provides a meaningful advance in stochastic CBF theory by bridging the gap between tight probabilistic guarantees and computational tractability. The explicit use of sub-Gaussian concentration to justify particle approximations is a clear strength, offering a more principled alternative to purely empirical risk measures in safety-critical control.

major comments (1)

[Section 4 (Finite-sample CVaR bounds and optimization formulation)] The finite-sample bounds on the CVaR approximation error (derived after the sub-Gaussian preservation result) are stated for any fixed control input u. Because the subsequent optimization is performed over u, an adversarial choice of u with respect to the realized particle set could inflate the approximation error beyond the stated bound. No covering-number, union-bound, or Lipschitz-continuity argument for the CVaR map with respect to u is visible in the derivation of the tractable optimization or the safety-certificate claim. This directly affects the central assertion that the resulting controller carries finite-sample probabilistic safety guarantees.

minor comments (3)

[Section 3.1] The notation for the sub-Gaussian parameter of the barrier increment should be introduced once with an explicit dependence on the Lipschitz constant and noise variance; subsequent uses are occasionally ambiguous.
[Section 5] Figure 3 (numerical results) would benefit from an additional panel or table reporting the empirical violation rate of the probabilistic constraint across Monte-Carlo trials, to allow direct visual comparison with the theoretical finite-sample bound.
[Section 4.3] A short remark on how the particle count N scales with the desired confidence level and the sub-Gaussian parameter would help readers assess practical implementation cost.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their careful review and for identifying this important technical consideration regarding uniformity of the finite-sample bounds. We address the comment directly below and will revise the manuscript accordingly.

read point-by-point responses

Referee: [Section 4 (Finite-sample CVaR bounds and optimization formulation)] The finite-sample bounds on the CVaR approximation error (derived after the sub-Gaussian preservation result) are stated for any fixed control input u. Because the subsequent optimization is performed over u, an adversarial choice of u with respect to the realized particle set could inflate the approximation error beyond the stated bound. No covering-number, union-bound, or Lipschitz-continuity argument for the CVaR map with respect to u is visible in the derivation of the tractable optimization or the safety-certificate claim. This directly affects the central assertion that the resulting controller carries finite-sample probabilistic safety guarantees.

Authors: We agree that the finite-sample CVaR error bounds in Section 4 are stated for fixed u and that the subsequent optimization over u requires a uniform guarantee to preserve the safety certificates. Under the manuscript's standing assumption of Lipschitz-continuous control-affine dynamics, the barrier increment is Lipschitz in u (with the sub-Gaussian parameter controlled uniformly on compact control sets). This implies Lipschitz continuity of the CVaR functional with respect to u. A standard epsilon-net argument over a compact control domain, followed by a union bound over the net, therefore yields a uniform finite-sample bound that holds with high probability over the particles independently of the optimized u. We will add this covering-number derivation explicitly to Section 4 in the revision (including any required compactness assumption on the control set), thereby ensuring the finite-sample probabilistic safety guarantees apply to the data-dependent controller. revision: yes

Circularity Check

0 steps flagged

No circularity; central claims derived from standard concentration inequalities and CBF theory

full rationale

The paper's derivation begins with the claim that Gaussian state-estimation noise propagating through Lipschitz control-affine dynamics preserves sub-Gaussianity of the barrier increment, which follows directly from the definition of sub-Gaussian random variables and the Lipschitz property without any self-referential fitting or redefinition. Finite-sample bounds on the particle CVaR approximation error are then obtained by applying standard sub-Gaussian tail inequalities (e.g., Hoeffding-type) to the empirical average over particles; these bounds are stated for fixed control inputs and do not rely on any parameter fitted to the target safety certificate itself. The subsequent tractable optimization simply substitutes the derived bounds into the CBF constraint, preserving the external mathematical grounding. No load-bearing step reduces to a self-citation, an ansatz smuggled from prior work by the same authors, or a renaming of an empirical pattern. The derivation is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on the assumption of Gaussian state estimation noise and Lipschitz-continuous control-affine dynamics; no free parameters or invented entities are mentioned in the abstract.

axioms (2)

domain assumption State estimation uncertainties are Gaussian.
Invoked to establish sub-Gaussianity of the barrier function increment.
domain assumption Dynamics are Lipschitz-continuous and control-affine.
Required for the propagation result that preserves sub-Gaussian structure.

pith-pipeline@v0.9.0 · 5487 in / 1282 out tokens · 48746 ms · 2026-05-10T18:14:58.203811+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

We establish that Gaussian uncertainties propagating through Lipschitz-continuous control-affine dynamics preserve sub-Gaussianity of the barrier function increment, with explicit tail bounds... finite-sample bounds on the approximation error between particle-based CVaR estimates
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Theorem 1 (Sub-Gaussianity of Δh)... σ_Δh ≤ C √(L_Φ,x² λ_max(Σ_x) + L_Φ,d² λ_max(Σ_d))

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

26 extracted references · 26 canonical work pages

[1]

Control barrier function based quadratic programs for safety critical systems,

A. D. Ames, X. Xu, J. W. Grizzle, and P. Tabuada, “Control barrier function based quadratic programs for safety critical systems,”IEEE Trans. Autom. Control, vol. 62, no. 8, pp. 3861–3876, 2017

work page 2017
[2]

Autonomous exploration of small bodies toward greater autonomy for deep space missions,

I. A. D. Nesnaset al., “Autonomous exploration of small bodies toward greater autonomy for deep space missions,”Frontiers in Robotics and AI, vol. 8, no. 11, pp. 1872–1873, 2021

work page 2021
[3]

Autonomy in the real-world: Au- tonomous trajectory planning for asteroid reconnaissance via stochas- tic optimization,

K. Echigo, A. Cauligi, S. Bandyopadhyay, D. Scharf, G. Lantoine, B. Acikmese, and I. Nesnas, “Autonomy in the real-world: Au- tonomous trajectory planning for asteroid reconnaissance via stochas- tic optimization,” inAIAA SciTech Forum, 2025

work page 2025
[4]

Robust safety under stochastic uncertainty with discrete-time control barrier functions,

R. K. Cosner, P. Culbertson, A. J. Taylor, and A. D. Ames, “Robust safety under stochastic uncertainty with discrete-time control barrier functions,” inProc. Robot. Sci. Syst. (RSS), 2023

work page 2023
[5]

Control barrier functions for complete and incomplete information stochastic systems,

A. Clark, “Control barrier functions for complete and incomplete information stochastic systems,” inProc. Amer. Control Conf. (ACC), 2019, pp. 2928–2935

work page 2019
[6]

arXiv preprint arXiv:2510.01501 , year=

P. Mestres, B. Werner, R. K. Cosner, and A. D. Ames. (2025) Proba- bilistic control barrier functions: Safety in probability for discrete-time stochastic systems. Available at https://arxiv.org/abs/2510.01501

work page arXiv 2025
[7]

Control barrier functions: Theory and applications,

A. D. Ames, S. Coogan, M. Egerstedt, G. Notomista, K. Sreenath, and P. Tabuada, “Control barrier functions: Theory and applications,” inProc. Eur. Control Conf. (ECC), 2019, pp. 3420–3431

work page 2019
[8]

Input-to-state safety with control barrier functions,

S. Kolathaya and A. D. Ames, “Input-to-state safety with control barrier functions,”IEEE Control Syst. Lett., vol. 3, no. 1, pp. 108– 113, 2019

work page 2019
[9]

Safe controller synthesis with tunable input-to-state safe control barrier functions,

A. Anil, A. J. Taylor, C. R. He, G. Orosz, and A. D. Ames, “Safe controller synthesis with tunable input-to-state safe control barrier functions,”IEEE Control Syst. Lett., vol. 6, pp. 908–913, 2022

work page 2022
[10]

Robust control barrier functions for constrained sta- bilization of nonlinear systems,

M. Jankovic, “Robust control barrier functions for constrained sta- bilization of nonlinear systems,”Automatica, vol. 96, pp. 359–367, 2018

work page 2018
[11]

Disturbance observer-based robust control barrier functions,

W. Wang and X. Xu, “Disturbance observer-based robust control barrier functions,” inProc. Amer. Control Conf. (ACC), 2023, pp. 3681–3687

work page 2023
[12]

Verification and control for finite-time safety of stochastic systems via barrier functions,

C. Santoyo, M. Dutreix, and S. Coogan, “Verification and control for finite-time safety of stochastic systems via barrier functions,” inProc. IEEE Conf. Control Technol. Appl., 2019, pp. 712–717

work page 2019
[13]

Probabilistically safe controllers based on control barrier functions and scenario model predictive control,

A. A. D. Nascimiento, A. Papachristodoulou, and K. Margellos, “Probabilistically safe controllers based on control barrier functions and scenario model predictive control,” inProc. IEEE Conf. Decis. Control (CDC), 2024, pp. 1814–1819

work page 2024
[14]

Control barrier functions for stochastic systems,

A. Clark, “Control barrier functions for stochastic systems,”Automat- ica, vol. 130, p. 109688, 2021

work page 2021
[15]

Risk-bounded control with Kalman filtering and stochastic barrier functions,

S. Yaghoubi, G. Fainekos, T. Yamaguchi, D. Prokhorov, and B. Hoxha, “Risk-bounded control with Kalman filtering and stochastic barrier functions,” inProc. IEEE Conf. Decis. Control (CDC), 2021, pp. 5213–5219

work page 2021
[16]

Finite-time regional verification of stochastic non-linear systems,

J. Steinhardt and R. Tedrake, “Finite-time regional verification of stochastic non-linear systems,”The International Journal of Robotics Research, vol. 31, no. 7, pp. 901–923, 2012

work page 2012
[17]

Belief control barrier functions for risk-aware control,

M. Vahs, C. Pek, and J. Tumova, “Belief control barrier functions for risk-aware control,”IEEE Robotics and Automation Letters, vol. 8, no. 12, pp. 8565–8572, 2023

work page 2023
[18]

Risk-aware control of discrete-time stochastic systems: Integrating Kalman filter and worst-case CVaR in control barrier functions,

M. Kishida, “Risk-aware control of discrete-time stochastic systems: Integrating Kalman filter and worst-case CVaR in control barrier functions,” inProc. IEEE Conf. Decis. Control (CDC), 2024, pp. 2019–2024

work page 2024
[19]

Risk-aware control for robots with non- Gaussian belief spaces,

M. Vahs and J. Tumova, “Risk-aware control for robots with non- Gaussian belief spaces,” inProc. IEEE Int. Conf. Robot. Autom. (ICRA), 2024, pp. 11 661–11 667

work page 2024
[20]

Discrete control barrier functions for safety-critical control of discrete systems with application to bipedal robot navigation,

A. Agrawal and K. Sreenath, “Discrete control barrier functions for safety-critical control of discrete systems with application to bipedal robot navigation,” inProc. Robot. Sci. Syst. (RSS), 2017, pp. 1–10

work page 2017
[21]

Vershynin,High-Dimensional Probability: An Introduction with Applications in Data Science

R. Vershynin,High-Dimensional Probability: An Introduction with Applications in Data Science. Cambridge University Press, 2018

work page 2018
[22]

The tight constant in the Dvoretzky-Kiefer-Wolfowitz inequality,

P. Massart, “The tight constant in the Dvoretzky-Kiefer-Wolfowitz inequality,”The Annals of Probability, vol. 18, no. 3, pp. 1269 – 1283, 1990

work page 1990
[23]

Concentration inequalities for conditional value at risk,

P. Thomas and E. Learned-Miller, “Concentration inequalities for conditional value at risk,” inProc. Int. Conf. Mach. Learn., 2019, pp. 6225–6233

work page 2019
[24]

Y . Zhao, X. Yu, M. Sesia, J. V . Deshmukh, and L. Lindemann. (2025) Conformal predictive programming for chance constrained optimization. Available at https://arxiv.org/abs/2402.07407

work page arXiv 2025
[25]

Optimization of conditional value- at-risk,

R. T. Rockafellar and S. Uryasev, “Optimization of conditional value- at-risk,”J. Risk, vol. 2, no. 3, pp. 21–41, 2000

work page 2000
[26]

Hybrid nonsmooth barrier functions with applications to provably safe and composable collision avoidance for robotic systems,

P. Glotfelter, I. Buckley, and M. Egerstedt, “Hybrid nonsmooth barrier functions with applications to provably safe and composable collision avoidance for robotic systems,”IEEE Robot. Autom. Lett., vol. 4, no. 2, pp. 1303–1310, 2019

work page 2019