cbf-cartpole

Introduction

Following repository is a minimalistic environment for experimenting with Control Barrier Functions (CBF).

The classical benchmark cartpole is used as the plant model. To keep it simple no simulator is used; instead, full non-linear dynamics of the cartpole is integrated.

Folder structure (ordered with increasing difficulty):

• acc.py: Most well-known, simple example: adaptive cruise control via CBF-QP. (not related to cartpole)

• cartpole.py: Cartpole dynamics and simulation, all controllers use it as plant

• nominal_control.py: Nominal controllers

• cbf_qp.py: Safety-Critical control with classical CBF, constrain cart velocity

CBF Enabled	CBF Disabled

• ecbf_qp.py: Exponential CBF with known model, constrain cart position
• cbf_qp_learning: Safety-Critical control with model uncertainty, CBF + learning (TODO)
• ecbf_qp_learning: Exponential CBF with unknown model, ECBF + learning (TODO)
• cbf_qp_id.py: CBF-QP embedded into inverse dynamics formulation.

Installation

1. Clone the repository

git clone https://github.com/Berk-Tosun/cbf-cartpole

2. Install required packages

pip install -r requirements.txt

Getting Started

Most python files in this project have a main guard. It means you can directly run them, e.g.:

python cbf_qp.py

This will generate one of the figures given in the introduction section.

What is CBF?

• Recommended introductory presentation: Jason Choi, UC Berkeley. Matching repository of the talk, where you can get the slides and the mentioned matlab code.

• Presentation by the main author of the CBF framework, Prof. Aaron Ames.

• Another introductory presentation: Air Lab Summer School 2020.

• Sample hardware application: Multi-Layered Safety for Legged Robots via Control Barrier Functions and Model Predictive Control. Its presentation from ICRA 2021.