Direct Perception of Easily Visible Information for Unknown Object Grasping

摘要

This paper discusses the estimation of the easily visible information for deciding action of unknown object grasping without all of predefined knowledge in a real situation. To estimate the easily visible information, active perception is an important concept. A conventional active perception method estimates an optimum sensing position to measure the accurate object information. However, a robot has to move all around the object to measure the accurate data beforehand. This is not realistic for a robot operation. On the other hand, a robot wants to know the most easily visible sensing position in the neighborhood of a current position to make a decision as soon as possible from a viewpoint of a robot action. As a first step, we propose an estimation of an easily visible information using a point cloud data of an object which is limited data. Now, we apply the unknown object detection method based on plane detection. And, we propose the graspability and the easily visible information estimation method from the point cloud data of the object. We verify the proposed method in a multiple object environment. The effectiveness of our proposal are discussed for grasping an unknown object.