I keep here my (old) notebooks on Model Predictive Control for path-following problems. Includes a Pybullet simulation to demo the controller. This mainly uses CVXPY as a framework. This repo contains code from other projecs, check them out in the special thanks section.
- To run the pybullet demo:
python3 mpc_demo_pybullet.py
- To run the simulation-less demo (simpler demo that does not use pybullet, useful for debugging):
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 class.
-
State space model derivation -> analytical and numerical derivaion of the model
-
MPC -> implementation and testing of various tweaks/improvements
-
Obstacle Avoidance -> Using halfplane constrains to avaoid track collisions -> Sill work in progress!
Racing car model is from: https://github.com/erwincoumans/pybullet_robots.
Results:
The environment can be repoduced via conda:
conda env create -f env.yml
conda activate simulation
The dependencies for just the python scripts can also be installed using pip
:
pip3 install --user --requirement requirements.txt