Probabilistic Safety under Arbitrary Disturbance Distributions using Piecewise-Affine Control Barrier Functions

Project page | Paper

---

This repository contains the safety filter implementations and experiments corresponding to the results in Section V of the paper "Probabilistic Safety under Arbitrary Disturbance Distributions using Piecewise-Affine Control Barrier Functions" by Matisse Teuwen, Mathijs Schuurmans and Panagiotis Patrinos.

Installation

Create a Python virtual environment (venv or conda). Clone the repository, and from the root, run

pip install .

to install our library. Reproducing the experiments from the paper can be done using the scripts in the experiments directory as explained below.

Reproducing the experiments

All experiments from the paper can be reproduced using the scripts in ./experiments.

Corridor Experiments (Section V-A)

Script	Section	Reproduces
corridor_known_distribution.py	V-A-1	Table I (empirical exit probabilities)
corridor_infeasibility.py	V-A-1	Figure 2 (σ–ε feasibility map)
corridor_unknown_distribution.py	V-A-2	Table II (comparison with scenario and conformal methods)
corridor_different_distributions.py	V-A-2	Figure 3 (comparison different noise distributions (Gaussian, Laplace, Student-t))

---

Path-Planning Experiments (Section V-B)

Script	Reproduces
path_planning_plot.py	Figure 4 (path planning example)
path_planning_timings.py	Timing statistics of Algorithm 1 vs. MIQP
miqp_solution_quality_compare.py	Suboptimality of Algorithm 1 vs. MIQP

Citation

If you use this code, please cite:

@article{teuwen2025-pwa-cbf,
  title         = "Probabilistic safety under arbitrary disturbance
                   distributions using piecewise-affine control barrier
                   functions",
  author        = "Teuwen, Matisse and Schuurmans, Mathijs and Patrinos,
                   Panagiotis",
  month         =  dec,
  year          =  2025,
  archivePrefix = "arXiv",
  primaryClass  = "math.OC",
  eprint        = "2512.04194"
}