In this project, we implement the Bayesian Optimization Algorithm (BOA) for autotuning the hyperparameters of Model Predictive Path Integral (MPPI) control to solve a planar box pushing task with non-trivial obstacles.
- Install all necessary dependencies using
bash install.sh. - Run our demo using
python demo.py. Instructions and results will be printed in the terminal and a live pushing demo should show up in pybullet GUI.
This implementation provides an easy-to-use Bayesian optimizer with full GPU pipeline support. The optimization results are at least comparable to the most popular BOA implementation in python.
Check here for more details.
- Python >= 3.8
- PyTorch >= 1.11
- GPytorch >= 1.9.1
BOA
f: objective function
X: support of variables
S: acquisition function
M: GP model
D: dataset
x: variable
y: observation
MPPI
