Gaussian Mixture Model

Using Sylvian Calinon's pbdlib repo to recreate GMM/HMM for Robot trajectories

The repository contains the procedure to recreate the GMM/HMM/HSMM models demonstrated in Sylvian Calinon's pbdlib-python repository [Link: https://gitlab.idiap.ch/rli/pbdlib-python.git] in order to train the robot trajectories for the aim of achieving LfD learning from demonstrations.

Procedure

Create Folder

Clone the repository

git clone https://gitlab.idiap.ch/rli/pbdlib-python.git

Setup Environment

Within the pbdlib-python folder setup the python 2.7 virtual environment using:

virtualenv -p /usr/bin/python2,7 py27venv

Enter the virtualenv using:

source py27venv/bin/activate

Install pbdlib

pip install -e .

Now copy the jupyter notebook ipynb file in this repository into the notebooks folder and open the jupyter notebook using

jupyter notebook notebooks/

Trajectories in the RobotData Folder

You will need the new_traj.mat matlab file in the RobotData folder within the /pbdlib/data/RobootData folder for this code to work.

Now all your trajectories are read and displayed in the jupyter file.

Stored values in files

Navigating to the folder /home/hulk/Documents/waypoints/ you will find two files f1.txt and f2.txt.

These files contain the saved values of the SIGMA and MU values as well as each trajectory points.

Check it out to plot the Gaussians and the reconstructed trajectory by editing the values in the Display_Gaussians.m file. This file uses the plot_gaussian_ellipsoid.m file also enclosed in the repo

This work is a combined effort of the Computer Vision and Robotics Laborartory at the University of Alberta under the supervision of Dr. Martin Jagersand.