Unknown object modeling on the basis of vision and pushing manipulation

Abstract

The authors aim at modeling unknown objects in human life environment using robot technologies. In this paper, the authors propose a method for modeling an unknown object on a desk using vision sensors, a manipulator, and a tactile sensor. The created model of an unknown object consists of the whole images and the 3-D point cloud with the texture. For getting the whole images, the unknown object is rotated by graspless manipulation. Pushing manipulation is used for the rotation. Pushing manipulation enables one to create models for objects that cannot be grasped by a robot hand. Tactile information is also used to confirm the pushing manipulation. The authors propose a method that decides a contact point and pushing direction for rotating the object from the fragmentary visual information. During the pushing manipulation, the object shape and the motion are measured by using a 3-D range camera and a stereo camera. The motion is estimated using 3-D flow. The model is built by combining the estimated motion and the measured 3-D shape. The method is useful for gathering objects' information in human life environment.