94 lines
3.0 KiB
Markdown
94 lines
3.0 KiB
Markdown
# mpc_python
|
|
|
|
I keep here my Jupyter notebooks on R&D on Model Predictive Control applyied to path-following problems in driverless vehicles. Includes also a Pybullet simulation to showcase the results.
|
|
This mainly uses **[CVXPY](https://www.cvxpy.org/)** as a framework. This repo contains code from other projecs, check them out in the special thanks section.
|
|
|
|
## Contents
|
|
|
|
### Usage
|
|
|
|
From within the `mpc_pybullet_demo` directory:
|
|
|
|
* To run the pybullet demo:
|
|
```bash
|
|
python3 mpc_demo_pybullet.py
|
|
```
|
|
|
|
* To run the simulation-less demo (simpler demo that does not use pybullet, useful for debugging):
|
|
```bash
|
|
python3 mpc_demo_pybullet.py
|
|
```
|
|
|
|
In both cases the script will promt the user for `enter` before starting the demo.
|
|
|
|
The settings for tuning the MPC controller are in the **[mpc_config](./mpc_pybullet_demo/mpcpy/mpc_config.py)** class.
|
|
|
|
### Jupyter Notebooks
|
|
|
|
1. State space model derivation -> analytical and numerical derivaion of the model
|
|
|
|
2. MPC -> implementation and testing of various tweaks/improvements
|
|
|
|
3. Obstacle Avoidance -> Using halfplane constrains to avaoid track collisions -> Sill **work in progress**!
|
|
|
|
<!--nobody cares about this
|
|
## About
|
|
|
|
The MPC is a model predictive path following controller which does follow a predefined reference by solving an optimization problem. The resulting optimization problem is shown in the following equation:
|
|
|
|
|
|
|
|
The terns of the cost function are the sum of the **reference tracking error**, **heading effort** and **actuaction rate of change**.
|
|
|
|
Where R,P,Q are the cost matrices used to tune the response.
|
|
|
|
The vehicle model is described by the bicycle kinematics model using the state space matrices A and B:
|
|
|
|
|
|
|
|
The state variables **(x)** of the model are:
|
|
|
|
* **x** coordinate of the robot
|
|
* **y** coordinate of the robot
|
|
* **v** velocuty of the robot
|
|
* **theta** heading of the robot
|
|
|
|
The inputs **(u)** of the model are:
|
|
|
|
* **a** linear acceleration of the robot
|
|
* **delta** steering angle of the robot
|
|
-->
|
|
|
|
### Results
|
|
|
|
Racing car model is from: *https://github.com/erwincoumans/pybullet_robots*.
|
|
|
|
|
|
|
|
Results:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
### Requirements
|
|
|
|
The environment can be repoduced via [conda](https://www.anaconda.com/products/distribution):
|
|
```bash
|
|
conda env create -f env.yml
|
|
conda activate simulation
|
|
```
|
|
|
|
The dependencies for just the python scripts can also be installed using `pip`:
|
|
```bash
|
|
pip3 install --user --requirement requirements.txt
|
|
```
|
|
|
|
## References & Special Thanks :star: :
|
|
* [Prof. Borrelli - mpc papers and material](https://borrelli.me.berkeley.edu/pdfpub/IV_KinematicMPC_jason.pdf)
|
|
* [AtsushiSakai - pythonrobotics](https://github.com/AtsushiSakai/PythonRobotics/)
|
|
* [erwincoumans - pybullet](https://pybullet.org/wordpress/)
|
|
* [alexliniger - mpcc](https://github.com/alexliniger/MPCC) and his [paper](https://onlinelibrary.wiley.com/doi/abs/10.1002/oca.2123)
|
|
* [arex18 - rocket-lander](https://github.com/arex18/rocket-lander)
|