[Paper Review] Almost-Sure Safety Guarantees of Stochastic Zero-Control Barrier Functions Do Not Hold

Oswin So, Andrew Clark+1 more|arXiv (Cornell University)|Dec 5, 2023

Insurance, Mortality, Demography, Risk Management4 citations

TL;DR

This paper demonstrates that the almost-sure safety guarantees claimed for stochastic zero-control barrier functions (ZCBF) in a 2021 paper are invalid, as shown by a counterexample using uncontrolled Brownian motion. The authors identify a flaw in the original proof's inductive argument and propose a corrected ZCBF condition using a modified proof technique, while also proving that stochastic reciprocal CBFs do provide almost-sure safety under appropriate drift-diffusion rate conditions.

ABSTRACT

The 2021 paper "Control barrier functions for stochastic systems" provides theorems that give almost sure safety guarantees given stochastic zero control barrier function (ZCBF). Unfortunately, both the theorem and its proof is invalid. In this letter, we illustrate on a toy example that the almost sure safety guarantees for stochastic ZCBF do not hold and explain why the proof is flawed. Although stochastic reciprocal barrier functions (RCBF) also uses the same proof technique, we provide a different proof technique that verifies that stochastic RCBFs are indeed safe with probability one. Using the RCBF, we derive a modified ZCBF condition that guarantees safety with probability one. Finally, we provide some discussion on the role of unbounded controls in the almost-sure safety guarantees of RCBFs, and show that the rate of divergence of the ratio of the drift and diffusion is the key for whether a system has almost sure safety guarantees.

Motivation & Objective

• To challenge the validity of almost-sure safety guarantees claimed for stochastic zero-control barrier functions (ZCBF) in a widely cited 2021 paper.
• To identify and explain the flaw in the inductive proof technique used in the original theorem for ZCBF safety.
• To demonstrate via a counterexample that the original ZCBF condition fails to ensure almost-sure safety for uncontrolled Brownian motion.
• To re-prove the almost-sure safety of stochastic reciprocal CBFs (RCBF) using a corrected proof technique.
• To derive a modified ZCBF condition that guarantees almost-sure safety by analyzing the divergence rate of the drift-to-diffusion ratio.

Proposed method

• Construct a counterexample using uncontrolled Brownian motion with a linear ZCBF $ h(x) = x $, showing that $ \Pr(W_t \geq 0 \, \forall t \geq 0) \neq 1 $, contradicting the original theorem.
• Identify the flaw in the original proof: an incorrect application of mathematical induction that assumes the stopping time sequence covers $[0, \infty)$ without proving $ \lim_{i\to\infty} \eta_i = \infty $.
• Use scale and speed measures from diffusion processes to analyze first-passage times and determine conditions under which absorption at the boundary occurs with probability one.
• Re-derive almost-sure safety for stochastic RCBFs using a new proof technique based on the scale function and the Feller test for explosions.
• Establish a modified ZCBF condition by analyzing the asymptotic behavior of the ratio of drift to diffusion coefficients, showing that if this ratio diverges sufficiently fast, almost-sure safety holds.
• Apply the Feller test for explosions to determine whether the process hits the boundary in finite time with positive probability, thereby characterizing the conditions for almost-sure safety.

Experimental results

Research questions

• RQ1 Does the original theorem in [5, Theorem 3] on almost-sure safety guarantees for stochastic ZCBFs hold under its stated conditions?
• RQ2 What is the flaw in the inductive proof technique used in the original proof of Theorem 3?
• RQ3 Can stochastic reciprocal CBFs (RCBFs) still guarantee almost-sure safety despite the flawed proof technique used for ZCBFs?
• RQ4 What conditions on the drift and diffusion coefficients ensure almost-sure safety for stochastic control systems?
• RQ5 Is there a modified ZCBF condition that restores almost-sure safety guarantees, and what is its mathematical basis?

Key findings

• The almost-sure safety guarantee for stochastic ZCBFs as stated in [5, Theorem 3] is invalid, as demonstrated by a counterexample with uncontrolled Brownian motion.
• The original proof fails because it assumes the union of stopping time intervals covers $[0, \infty)$ without proving $ \lim_{i\to\infty} \eta_i = \infty $, invalidating the inductive step.
• For stochastic RCBFs, the same proof technique fails, but a new proof technique based on scale functions and the Feller test confirms that RCBFs do guarantee almost-sure safety.
• A modified ZCBF condition is derived that ensures almost-sure safety by requiring the ratio of drift to diffusion to diverge sufficiently fast as the state approaches the boundary.
• The key factor for almost-sure safety is the rate of divergence of the drift-to-diffusion ratio: if this ratio grows fast enough, the process almost surely avoids the unsafe region.
• When the drift-to-diffusion ratio diverges slowly or is bounded, the process may hit the boundary in finite time with positive probability, invalidating almost-sure safety.