Deep Object Pose Estimation for Semantic Robotic Grasping of Household\n Objects

Abstract

Using synthetic data for training deep neural networks for robotic\nmanipulation holds the promise of an almost unlimited amount of pre-labeled\ntraining data, generated safely out of harm's way. One of the key challenges of\nsynthetic data, to date, has been to bridge the so-called reality gap, so that\nnetworks trained on synthetic data operate correctly when exposed to real-world\ndata. We explore the reality gap in the context of 6-DoF pose estimation of\nknown objects from a single RGB image. We show that for this problem the\nreality gap can be successfully spanned by a simple combination of domain\nrandomized and photorealistic data. Using synthetic data generated in this\nmanner, we introduce a one-shot deep neural network that is able to perform\ncompetitively against a state-of-the-art network trained on a combination of\nreal and synthetic data. To our knowledge, this is the first deep network\ntrained only on synthetic data that is able to achieve state-of-the-art\nperformance on 6-DoF object pose estimation. Our network also generalizes\nbetter to novel environments including extreme lighting conditions, for which\nwe show qualitative results. Using this network we demonstrate a real-time\nsystem estimating object poses with sufficient accuracy for real-world semantic\ngrasping of known household objects in clutter by a real robot.\n