Using orthogonal basis functions and template matching to learn whiteboard cleaning task by imitation

Abstract

In this paper we present an integrated approach to learn complex trajectories by imitation. A complex trajectory is a trajectory which includes some simple sub-trajectories. In this method Orthogonal basis Function and Template Matching (OFTM) is used besides Gaussian Mixture Model (GMM) to learn new trajectories more efficient in both speed and accuracy. This method is implemented on a three DOF robot arm to learn complex trajectories which are related to whiteboard cleaning task. In this method, the robot uses primitive movements including template and orthogonal basis trajectories, which are learnt by using Gaussian Mixture Model (GMM), to construct new given trajectories. To obtain this goal, the robot calculates the dissimilarity between the new trajectory and arbitrary templates, then the similar parts will be replaced by the template, and the rest of the new trajectory will be constructed by using the orthogonal learnt trajectories. The results show that our method is more accurate and requires less computation in comparison with learning the whole trajectory by GMM.