Manipulative Tasks Identification by Learning and Generalizing Hand Motions

Abstract

In this work is proposed an approach to learn patterns and recognize a manipulative task by the extracted features among multiples observations. The diversity of information such as hand motion, fingers flexure and object trajectory are important to represent a manipulative task. By using the relevant features is possible to generate a general form of the signals that represents a specific dataset of trials. The hand motion generalization process is achieved by polynomial regression. Later, given a new observation, it is performed a classification and identification of a task by using the learned features.