StochasticBarrier.jl: A Toolbox for Stochastic Barrier Function Synthesis

A research team from institutions including the University of Colorado Boulder has released StochasticBarrier.jl, a groundbreaking open-source toolbox for the formal safety verification of stochastic AI and control systems. The tool, written in Julia, synthesizes Stochastic Barrier Functions (SBFs)—mathematical certificates that guarantee a system will remain within safe operating limits despite random noise. It represents a significant leap in capability, certifying safety for a broad class of systems including linear, polynomial, and crucially, piecewise affine (PWA) dynamics, which can approximate complex, nonlinear behaviors common in modern AI agents and robotics.

The toolbox implements two core methodologies: a Sum-of-Squares (SOS) optimization approach using semi-definite programming solvers, and a novel method based on piecewise constant (PWC) functions with engines using linear programming and gradient descent. In rigorous benchmarking against over 30 case studies, StochasticBarrier.jl demonstrated staggering performance gains. It was found to be up to four orders of magnitude (10,000x) faster than its closest competitor, while also providing tighter, more conservative bounds on safety probabilities and scaling effectively to verify higher-dimensional systems that were previously intractable. This breakthrough dramatically lowers the computational barrier to proving the real-world safety of autonomous systems, from drones to AI decision-making processes.