UniGrasp: Learning a Unified Model to Grasp With Multifingered Robotic Hands

Abstract

To achieve a successful grasp, gripper attributes including geometry and kinematics play a role equally important to the target object geometry. The majority of previous work has focused on developing grasp methods that generalize over novel object geometry but are specific to a certain robot hand. We propose UniGrasp, an efficient data-driven grasp synthesis method that considers both the object geometry and gripper attributes as inputs. UniGrasp is based on a novel deep neural network architecture that selects sets of contact points from the input point cloud of the object. The proposed model is trained on a large dataset to produce contact points that are in force closure and reachable by the robot hand. By using contact points as output, we can transfer between a diverse set of Nfingered robotic hands. Our model produces over 90% valid contact points in Top10 predictions in simulation and more than 90% successful grasps in the real world experiments for various known two-fingered and three-fingered grippers. Our model also achieves 93% and 83% successful grasps in the real world experiments for a novel two-fingered and five-fingered anthropomorphic robotic hand, respectively.

Robotics Dexterous Grasping: The Methods Based on Point Cloud and Deep Learning

“…Then RANSAC is utilized to acquire the orientation of the grasp from the pixel mask and point cloud. Instead of generating the grasp pose parameters, Shao et al ( 2020 ) attempts to predict the grasp contact points. It not only extracts the feature of object point cloud, but also pays attention to gripper properties.…”

“…Despite this, there are still some works focus on multi-finger grasping have achieved satisfactory results. Guan et al ( 2019 ), Lin and Cong ( 2019 ), Liu C. et al ( 2019 ), Wu et al ( 2019 ), Shao et al ( 2020 ), and Yu Y. et al ( 2019 ), Yu Y. et al ( 2020 ) utilize three-finger hand as their end effectors and Ficuciello et al ( 2019 ) and Yu Q. et al ( 2020 ) adopt five-finger to design grasp learning algorithm.…”

“…The methods mentioned in sections Grasping Candidate Generation and Grasp Candidate Evaluation based on CNN to generate and evaluate grasp candidates show the potentials of generalization (Mahler et al, 2017 , 2018 ; Yan et al, 2019 ; Chen et al, 2020 ; Shao et al, 2020 ). However, for the proposed approaches based on reinforcement learning are hard to be transferred to other tasks.…”

DDGC: Generative Deep Dexterous Grasping in Clutter

“…These methods generalize across many different objects but still rely on an inherently slow sampling process. Therefore, other deep-learning methods have focused on learning the actual grasp sampler either from raw sensor inputs [5], [7], [27]- [29] or shape completed models [8]. The limitation of all of these methods, with the exception of [7], is that they are proposed for and evaluated on single object grasping.…”

Emergent Hand Morphology and Control from Optimizing Robust Grasps of Diverse Objects

“…Meanwhile, recent advances in learning for robotics have yielded studies on data-driven techniques for robot grasping [3], [4]. Most of them have focused on grasp planning and selection [5], i.e., the control algorithm of grasping, with fixed and often simple parallel-jaw grippers. In contrast, customizing the robot embodiment to a particular task simplifies the computational burden of decision making, a phenomenon referred to as morphological computation in robot control [6].…”