This repository contains code for the paper "Lipschitz Safe Bayesian Optimization for Automotive Control" which has been accepted to the 2024 Conference for Decision and Control in Milano, Italy.
If you find our code or paper useful, please consider citing
@article{menn2024lipschitz,
title={Lipschitz Safe {Bayesian} Optimization for Automotive Control},
author={Menn, Johanna and Pelizzari, Pietro and Fleps-Dezasse, Michael and Trimpe, Sebastian},
booktitle={2024 63nd IEEE Conference on Decision and Control (CDC)},
year={2024},
organization={IEEE}
}
We propose a new algorithm MCLoSBO, that uses the safety mechanism of Lipschitz-only Safe Bayesian Optimization (LoSBO) in a multiple constraints setting.
Our implementation uses SafeOptMC as a basis and a baseline for the experiments.
The code in this repository can be used to reproduce the figures and results of our paper. You can reproduce the experiments by runnig the
python experiments.py
and reproduce the plot from the paper with
python evaluate_simulation_results.py
As this project was a collaboration with industry, we cannot provide the orginal simulator, that was used for the experiments in the paper. As a work around, we queried the simulator with a fine grid over the parameter space and evaluated the objective and constraint functions. The results of this gridding are used when running the experiments script.
Into an environment with Python 3.11 you can install the needed packages with
pip install -r requirements.txt