Teaching grasping points using natural movements

Abstract

The research on robots performing every-day tasks at home, has pursued the problem of the manipulation of everyday objects. Among them, grasping a cloth is a challenging task, as the textile is highly-deformable and it is not straightforward to define a generic grasping point. In this paper, we address this problem by introducing a new robot interaction method that enables unexperienced users to control the robot in a natural way. When the robot proposes a grasping point, the user is able to teach the robot a new one. The data collected using this method is then used for training a system using linear regression method, which produces better grasping points and allowing better manipulation actions. The experiments demonstrates the validity of the new interaction method and its potential to improve the point-grasping selection algorithm.