CaTGrasp: Learning Category-Level Task-Relevant Grasping in Clutter from Simulation

Abstract

Task-relevant grasping is critical for industrial assembly, where downstream manipulation tasks constrain the set of valid grasps. Learning how to perform this task, however, is challenging, since task-relevant grasp labels are hard to define and annotate. There is also yet no consensus on proper representations for modeling or off-the-shelf tools for performing task-relevant grasps. This work proposes a framework to learn task-relevant grasping for industrial objects without the need of time-consuming real-world data collection or manual annotation. To achieve this, the entire framework is trained solely in simulation, including supervised training with synthetic label generation and self-supervised, hand-object interaction. In the context of this framework, this paper proposes a novel, object-centric canonical representation at the category level, which allows establishing dense correspondence across object instances and transferring task-relevant grasps to novel instances. Extensive experiments on task-relevant grasping of densely-cluttered industrial objects are conducted in both simulation and real-world setups, demonstrating the effectiveness of the proposed framework. Code and data will be released upon acceptance at https://sites.google.com/view/catgrasp.

TransGrasp: Grasp Pose Estimation of a Category of Objects by Transferring Grasps from Only One Labeled Instance

“…DGCM-Net [19] transfers grasps by predicting view-dependent normalized object coordinate (VD-NOC) values between pairs of depth images. CaTGrasp [29] maps the input point cloud to a Non-Uniform Normalized Object Coordinate Space (NUNOCS) where the spatial correspondence is built. DON [9] learns pixel-wise descriptors in a self-supervised manner to transfer grasp points between 2D images.…”

Interleaving Monte Carlo Tree Search and Self-Supervised Learning for Object Retrieval in Clutter

“…These challenges have given rise to data-driven methods that learn from data, where many works focus on isolated objects [13]- [16]. Recently, grasping in clutter has received more attention [17]- [21]. Convolutional Neural Networks are widely used to construct grasp proposal networks such as Dex-net 4.0 [22], which are trained to detect 6D grasp poses in point clouds [23].…”