Instance Segmentation of Visible and Occluded Regions for Finding and\n Picking Target from a Pile of Objects

Abstract

We present a robotic system for picking a target from a pile of objects that\nis capable of finding and grasping the target object by removing obstacles in\nthe appropriate order. The fundamental idea is to segment instances with both\nvisible and occluded masks, which we call `instance occlusion segmentation'. To\nachieve this, we extend an existing instance segmentation model with a novel\n`relook' architecture, in which the model explicitly learns the inter-instance\nrelationship. Also, by using image synthesis, we make the system capable of\nhandling new objects without human annotations. The experimental results show\nthe effectiveness of the relook architecture when compared with a conventional\nmodel and of the image synthesis when compared to a human-annotated dataset. We\nalso demonstrate the capability of our system to achieve picking a target in a\ncluttered environment with a real robot.\n