Programming by Demonstration of Object Grasping Using Gaussian Mixture Model Based Techniques

Abstract

A programming by demonstration techniques combination that allows object grasping is presented. The techniques are the task-parameterized Gaussian mixture model and the Gaussian mixture models. With the first technique, a path hand to the new position and orientation of the object is estimated. Nevertheless, the final value obtained of position and orientation is not always accurate, so it encourages to apply a second method or step correction using Gaussian mixture models. A real experiment allow us to obtain a database of different grasps to subsequently validate the proposed algorithm. The obtained results by using a grasp quality subjective evaluation and graphics error show an improvement to the result generated only by using the task-parameterized Gaussian mixture model