Two-stage robotic picking for randomly stacked objects with recognition difficulty

Abstract

In this paper, we propose a novel two-stage approach on picking an object from randomly stacked pile. We especially consider a situation where it is difficult for a 3D depth sensor to identify the pose of objects such as shinny ones and black ones. For such objects, our method first roughly grasps some of the objects from the pile without using the visual information and roughly place them onto a working table. After some of the objects are placed on a working table, a robot picks one of them by detecting the 2D position of objects placed on a working table by using the 2D RGB image. We performed experiments for several objects with different shape and weight. Through experimental study, we confirmed that a robot can identify the position of one of the objects after the objects are placed on a working table and that one of the objects can be successfully picked up by a robot.