Update README.md

master
Jun Zeng 2020-12-20 19:26:06 -08:00
parent 4bda21769a
commit bee361feb9
3 changed files with 56 additions and 43 deletions

View File

@ -1,48 +1,28 @@
# MPC-CBF
Model Predictive Control with Control Barrier Functions
We propose a control framework which unifies the model predictive control and control barrier functions. This is the reference implementation of our paper:
### Safety-Critical Model Predictive Control with Discrete-Time Control Barrier Function
[PDF](https://arxiv.org/abs/2007.11718) | [Code: Double Integratror](double-integrator-2D) | [Code: Car Racing](car-racing)
![](car-racing/demo.gif)
*Jun Zeng, Bike Zhang and Koushil Sreenath*
# Prerequisites
## Matlab
<img src="car-racing/demo.gif" height="200"/>
#### Citing
If you find this code useful in your work, please consider citing:
```shell
@article{zeng2020mpccbf,
title={Safety-critical model predictive control with discrete-time control barrier function},
author={Zeng, Jun and Zhang, Bike and Sreenath, Koushil},
journal={arXiv preprint arXiv:2007.11718},
year={2020}
}
```
### Instructions
Two independent markdown files are written to illustrate numerical examples of [double integrator](double-integrator-2D/README.md) and [racing car](car-racing/README.md).
### Dependencies
#### Matlab
The packages needed for running the code are [Yalmip](https://yalmip.github.io/) and [IPOPT](https://projects.coin-or.org/Ipopt/wiki/MatlabInterface).
We also provide the zipped version of precompiled .mex files for IPOPT in the folder `packages` in case you don't have it. Unzip that file and add those .mex files into your MATLAB path.
# Description
## 2D double integrator
The 2D double integrator is assigned to reach the target position at origin while avoiding obstacles. We have three classes for different controllers: `DCLF_DCBF.m` (DCLF-DCBF), `MPC_CBF.m` (MPC-CBF) and `MPC_DC` (MPC-DC), respectively.
Moreover, to illustrate the performance among them, we have:
* `main.m`: Run DCLF-DCBF/MPC-CBF/MPC-DC respectively.
* `analysis_gamma.m`: Run analysis for different hyperparameter $\gamma$.
* `analysis_horizon.m`: Run analysis for different horizon.
We illustrate the performance between DCLF-DCBF/MPC-DC/MPC-CBF
| DCLF-DCBF | MPC-DC (N=8) |
| --- | --- |
| <img src="double-integrator-2D/figures/dclf-dcbf-avoidance.png" width="200" height="200"> | <img src="double-integrator-2D/figures/mpc-dc-avoidance.png" width="200" height="200"> |
| MPC-CBF (N=1) | MPC-CBF (N=8) |
| --- | --- |
| <img src="double-integrator-2D/figures/mpc-cbf-avoidance-one-step.png" width="200" height="200"> | <img src="double-integrator-2D/figures/mpc-cbf-avoidance-several-steps.png" width="200" height="200"> |
and also the safety performance for different numbers of horizon and hyperparameters
| Different hyperparameter | Different horizon |
| --- | --- |
| <img src="double-integrator-2D/figures/benchmark-gamma.png" width="200" height="200"> | <img src="double-integrator-2D/figures/benchmark-horizon.png" width="200" height="200"> |
## Car racing competition
We have the speed profile and control input shown as follow,
<img src="car-racing/lmpc-speed-norm-profile.png" width="400">
<img src="car-racing/lmpc-deviation-profile.png" width="400">
<img src="car-racing/lmpc-input-profile.png" width="400">
The animation can be found on the top of this readme, we will release full code after the paper is accepted.
# References
This repository is based on the following:
* Jun Zeng, Bike Zhang and Koushil Sreenath. "Safety-Critical Model Predictive Control with Discrete-Time Control Barrier Function." [[PDF]](http://arxiv.org/abs/2007.11718)

12
car-racing/README.md Normal file
View File

@ -0,0 +1,12 @@
### Car racing competition
The source codes are mainly adapted from Ugo's LMPC code and there are some lagacy features which are not used in our paper.
We simulate a car racing competition between several cars, and ego car's speed profile and control input are shown as follow,
<img src="lmpc-speed-norm-profile.png" width="400">
<img src="lmpc-deviation-profile.png" width="400">
<img src="lmpc-input-profile.png" width="400">
The animation can be found on the top of this readme, we will release full code after the paper is accepted.

View File

@ -0,0 +1,21 @@
### 2D double integrator
The 2D double integrator is assigned to reach the target position at origin while avoiding obstacles. We have three classes for different controllers: `DCLF_DCBF.m` (DCLF-DCBF), `MPC_CBF.m` (MPC-CBF) and `MPC_DC` (MPC-DC), respectively.
Moreover, to illustrate the performance among them, we have:
* `main.m`: Run DCLF-DCBF/MPC-CBF/MPC-DC respectively.
* `analysis_gamma.m`: Run analysis for different hyperparameter $\gamma$.
* `analysis_horizon.m`: Run analysis for different horizon.
We illustrate the performance between DCLF-DCBF/MPC-DC/MPC-CBF
| DCLF-DCBF | MPC-DC (N=8) |
| --- | --- |
| <img src="figures/dclf-dcbf-avoidance.png" width="200" height="200"> | <img src="figures/mpc-dc-avoidance.png" width="200" height="200"> |
| MPC-CBF (N=1) | MPC-CBF (N=8) |
| --- | --- |
| <img src="figures/mpc-cbf-avoidance-one-step.png" width="200" height="200"> | <img src="figures/mpc-cbf-avoidance-several-steps.png" width="200" height="200"> |
and also the safety performance for different numbers of horizon and hyperparameters
| Different hyperparameter | Different horizon |
| --- | --- |
| <img src="figures/benchmark-gamma.png" width="200" height="200"> | <img src="figures/benchmark-horizon.png" width="200" height="200">