Motion Planning to Retrieve an Object from Random Pile

Abstract

It is challenging to retrieve a target object from a randomly stacked pile by using a robot due to the occlusion of the target object. In this study, we propose a novel retrieval method in which a robot selects the viewpose to observe the occlusion part of the target object using the RGB-D images, and then selects the motion of grasping/dragging to retrieve the object depending on the configuration of the pile. We experimentally confirm that a robot effectively observes a pile with a complex configuration and successfully retrieves a target object.