/safe-exploration

Safe Exploration with MPC and Gaussian process models

Primary LanguagePythonMIT LicenseMIT

Learning-based Model Predictive Control for Safe Exploration

https://circleci.com/gh/befelix/safe-exploration/tree/master.svg?style=svg https://travis-ci.com/befelix/safe-exploration.svg?branch=master

This code accompanies the following paper:

[1]T. Koller, F. Berkenkamp, M. Turchetta, A. Krause, Learning-based Model Predictive Control for Safe Exploration in Proc. of the Conference on Decision and Control (CDC), 2018

Installation

Install the library including all dependencies with.

pip install -e ".[test,visualization,ssm_gpy,ssm_pytroch]"

test for the testing tools. visualization for visualizations such as matplotlib. ssm_gpy and ssm_pytorch for state space models based on GPy or PyTorch, respectively.

Experiments can be run using the experiments/run.py script.

Test can be run using pytest. There are also more sophisticated style tests in scripts/test_code.sh.