Try it online:

GMM-lbd

Gaussian mixture experiments for learning by demonstration

Notebook examples:

• example 1
• example 2

Installation

pip install gmm-lbd

OR if you clone this repository, you can make a

python setup.py develop

Roadmap

Low level (Gaussian mixture)

• Automaticly choose the number of gaussians
• conditional probability of GMM
• regression of GMM
• product of retrived means and covariances of GMM
• product with non consistent shapes
• speed management

High level (combinaison of movements)

• Quick add pypot records for any motors
• Easyly represent GMM with ellipses
• Plot ellipse for GMM with more than 2 dimensions
• Align movements with DTW
• White detection (at begin and end of movements)
• Sequential : concatenation of GMMs
• Concurent : product
• Add a coefficiant to rise or low the importance of a movement
• Add a filter in the sequential combinaison
• Adapt to use an IK model, for performing task space trajectory: WANTED
• Incremental definition of the GMM for each new representation @Calinon07HRI (very good idea for online learning)

Pypot records improvements (TODO)

• record datas with a variable framerate (compression + CPU usage for generating GMMs )