Institute for
Robotics and Process Control

Deutsch   English

Sorry, currently in German language only...

seminar topic

Gaussian Mixtures for movement learning

course: Bachelor/Master
A very popular model for imitation learning of movement are Gaussian Mixtures. In the basic version, they take data from human demonstration, model these through a standard sum of Gaussian functions and finally use Gaussian Mixture Regression to recover an approximate to the "mean movement", which then can be executed on a robot. Many versions of this simple idea have been derived and used to learn skills like stacking objects, put a spoon to a mouth, or grasp objects bimanually or on a chess board.

Literature: Calinon et al.(External)(PDF) Mühling et al.(External)(PDF)

zurück zur Übersicht

It took 0.15s to generate this page.