Deep Learning for Detecting Robotic Grasps

Lenz, Ian; Lee, Honglak; Saxena, Ashutosh

doi:10.15607/rss.2013.ix.012

Cited by 87 publications

(24 citation statements)

References 15 publications

(25 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Another line of research has focused on synthesizing grasps using statistical models [5] learned from a database of images [27] or point clouds [9], [15], [46] of objects annotated with grasps from human demonstrators [15], [27] or physical execution [15]. Kappler et al [18] created a database of over 700 object instances, each labelled with 500 Barrett hand grasps and their associated quality from human annotations and the results of simulations with the ODE physics engine.…”

Section: Related Workmentioning

confidence: 99%

See 1 more Smart Citation

Dex-Net 1.0: A cloud-based network of 3D objects for robust grasp planning using a Multi-Armed Bandit model with correlated rewards

Mahler

Pokorny

Hou

et al. 2016

2016 IEEE International Conference on Robotics and Automation (ICRA)

316

237

View full text Add to dashboard Cite

This paper presents the Dexterity Network (DexNet) 1.0, a dataset of 3D object models and a sampling-based planning algorithm to explore how Cloud Robotics can be used for robust grasp planning. The algorithm uses a MultiArmed Bandit model with correlated rewards to leverage prior grasps and 3D object models in a growing dataset that currently includes over 10,000 unique 3D object models and 2.5 million parallel-jaw grasps. Each grasp includes an estimate of the probability of force closure under uncertainty in object and gripper pose and friction. Dex-Net 1.0 uses Multi-View Convolutional Neural Networks (MV-CNNs), a new deep learning method for 3D object classification, to provide a similarity metric between objects, and the Google Cloud Platform to simultaneously run up to 1,500 virtual cores, reducing experiment runtime by up to three orders of magnitude. Experiments suggest that correlated bandit techniques can use a cloud-based network of object models to significantly reduce the number of samples required for robust grasp planning. We report on system sensitivity to variations in similarity metrics and in uncertainty in pose and friction. Code and updated information is available at http://berkeleyautomation.github.io/dex-net/.

show abstract

Section: Related Workmentioning

confidence: 99%

“…Can analogous scaling effects emerge when datasets of 3D object models are applied to learning robust grasping and manipulation policies for robots? This question is being explored by others [12], [18], [27], [32], [33], and in this paper we present initial results using a new dataset of 3D models and grasp planning algorithm.…”

Section: Introductionmentioning

confidence: 99%

Dex-Net 1.0: A cloud-based network of 3D objects for robust grasp planning using a Multi-Armed Bandit model with correlated rewards

Mahler

Pokorny

Hou

et al. 2016

2016 IEEE International Conference on Robotics and Automation (ICRA)

316

237

View full text Add to dashboard Cite

show abstract

“…Our robot always lifts the handle side of the screw driver without keeping it horizontal, as this grasp type doesn't have a sufficient wrench to grasp such objects whose CG is far away from the contacts. 'PS3 joystick' is a difficult object to grasp for us, which is also reported in [19]. There is only a little viable pinch grasps which can be used to lift the object.…”

Section: Object Modelling and Grasp Generationmentioning

confidence: 98%

Precision grasping based on probabilistic models of unknown objects

Chen

Dietrich

Wichert

2016

2016 IEEE International Conference on Robotics and Automation (ICRA)

View full text Add to dashboard Cite

Reliable precision grasping for unknown objects is a prerequisite for robots that work in the field of logistics, manufacturing and household tasks. The nature of this task requires a simultaneous solution of a mixture of sub-problems. These include estimating object properties, finding viable grasps and executing grasps without displacement. We propose to explicitly take perceptual uncertainty into account during grasp execution. The underlying object representation is a probabilistic signed distance field, which includes both signed distances to the surface and spatially interpretable variances. Based on this representation, we propose a two-stage grasp generation method, which is specifically designed for generating precision grasps. In order to evaluate the whole approach, we perform extensive real world grasping experiments on a set of hard-to-grasp objects. Our approach achieves 78% success rate and shows robustness to the placement orientation.

show abstract

“…Modeling and predicting grasping interactions received special attention in robotics [Bohg et al 2013]. Data-driven techniques often rely on machine learning and train on annotated example shapes to predict graspable regions based on their geometric features [Saxena et al 2006;Saxena 2009;Lenz et al 2013]. Alternatively, one can use shape retrieval to find a similar object in an annotated database and transfer a grasping pose [Goldfeder and Allen 2011].…”

Section: Related Workmentioning

confidence: 99%

Shape2Pose

2014

View full text Add to dashboard Cite

As 3D acquisition devices and modeling tools become widely available there is a growing need for automatic algorithms that analyze the semantics and functionality of digitized shapes. Most recent research has focused on analyzing geometric structures of shapes. Our work is motivated by the observation that a majority of manmade shapes are designed to be used by people. Thus, in order to fully understand their semantics, one needs to answer a fundamental question: "how do people interact with these objects?" As an initial step towards this goal, we offer a novel algorithm for automatically predicting a static pose that a person would need to adopt in order to use an object. Specifically, given an input 3D shape, the goal of our analysis is to predict a corresponding human pose, including contact points and kinematic parameters. This is especially challenging for man-made objects that commonly exhibit a lot of variance in their geometric structure. We address this challenge by observing that contact points usually share consistent local geometric features related to the anthropometric properties of corresponding parts and that human body is subject to kinematic constraints and priors. Accordingly, our method effectively combines local region classification and global kinematically-constrained search to successfully predict poses for various objects. We also evaluate our algorithm on six diverse collections of 3D polygonal models (chairs, gym equipment, cockpits, carts, bicycles, and bipedal devices) containing a total of 147 models. Finally, we demonstrate that the poses predicted by our algorithm can be used in several shape analysis problems, such as establishing correspondences between objects, detecting salient regions, finding informative viewpoints, and retrieving functionally-similar shapes.

show abstract

Deep Learning for Detecting Robotic Grasps

Cited by 87 publications

References 15 publications

Dex-Net 1.0: A cloud-based network of 3D objects for robust grasp planning using a Multi-Armed Bandit model with correlated rewards

Dex-Net 1.0: A cloud-based network of 3D objects for robust grasp planning using a Multi-Armed Bandit model with correlated rewards

Precision grasping based on probabilistic models of unknown objects

Shape2Pose

Contact Info

Product

Resources

About