Modeling Grasp Type Improves Learning-Based Grasp Planning

Lu, Qingkai; Hermans, Tucker

doi:10.1109/lra.2019.2893410

Cited by 35 publications

(33 citation statements)

References 36 publications

(69 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Hyttinen et al (2015) used tactile signatures fed to a trained classifier to predict object grasp stability. Lu and Hermans (2019) learned a probabilistic model of grasp success as a function from a reduced dimension summary of the sensory state and a specification of a pre-grasp configuration. This is conditioned on a prior distribution over grasp types.…”

Section: Learning a Grasp Evaluation Functionmentioning

confidence: 99%

Learning better generative models for dexterous, single-view grasping of novel objects

Kopicki

Belter

Wyatt

2019

The International Journal of Robotics Research

View full text Add to dashboard Cite

This paper concerns the problem of how to learn to grasp dexterously, so as to be able to then grasp novel objects seen only from a single view-point. Recently, progress has been made in data-efficient learning of generative grasp models which transfer well to novel objects. These generative grasp models are learned from demonstration (LfD). One weakness is that, as this paper shall show, grasp transfer under challenging single view conditions is unreliable. Second, the number of generative model elements rises linearly in the number of training examples. This, in turn, limits the potential of these generative models for generalisation and continual improvement. In this paper, it is shown how to address these problems. Several technical contributions are made: (i) a view-based model of a grasp; (ii) a method for combining and compressing multiple grasp models; (iii) a new way of evaluating contacts that is used both to generate and to score grasps. These, together, improve both grasp performance and reduce the number of models learned for grasp transfer. These advances, in turn, also allow the introduction of autonomous training, in which the robot learns from self-generated grasps. Evaluation on a challenging test set shows that, with innovations (i)-(iii) deployed, grasp transfer success rises from 55.1% to 81.6%. By adding autonomous training this rises to 87.8%. These differences are statistically significant. In total, across all experiments, 539 test grasps were executed on real objects.

show abstract

Section: Learning a Grasp Evaluation Functionmentioning

confidence: 99%

Learning better generative models for dexterous, single-view grasping of novel objects

Kopicki

Belter

Wyatt

2019

The International Journal of Robotics Research

View full text Add to dashboard Cite

show abstract

“…8,9,11 Traditional grasping control strategies without vision sensors are prone to leading to grasping failure when uncertainties occur in pose estimation. 12 Visual servoing is an effective method to improve success rate of grasping for the promising ability to provide noncontact online information. 9,11 Series of visual servoing control algorithms have been employed in grasping manipulation, which dramatically increases the flexibility and reliability.…”

Section: Introductionmentioning

confidence: 99%

Automatic precision robot assembly system with microscopic vision and force sensor

Zhou

et al. 2019

International Journal of Advanced Robotic Systems

View full text Add to dashboard Cite

An automatic precision robot assembly system is established. The robot assembly system mainly consists of an industrial robot, three cameras, a micro force sensor, and a specific gripper. The industrial robot is a six-axis serial manipulator, which is used to conduct grasping and assembly subtasks. Two microscopic cameras are fixed on two high accuracy translational platforms to provide visual information in aligning stage for assembly. While one conventional camera is installed on the robotic end effector to guide the gripper to grasp component. The micro force sensor is installed on the robotic end effector to perceive the contacted forces in inserting stage. According to the characteristics of components, an adsorptive gripper is designed to pick up components. In addition, a three-stage "aligning-approaching-grasping" control strategy for grasping subtask and a two-stage "aligning-inserting" control strategy for assembly subtask are proposed. Position offset compensation is computed and introduced into aligning stage for assembly to make the grasped component in the microscopic cameras' small field of view. Finally, based on the established robot assembly system and the proposed control strategies, the assembly tasks including grasping and assembly are carried out automatically. With 30 grasping experiments, the success rate is 100%. Besides, the position and orientation alignment errors of pose alignment for assembly are less than 20 mm and 0.1 .

show abstract

“…Various techniques of robot grasp planning [1], [2], and [3] have been achieved over the years. Grasp planning has been achieved using probabilistic inference [1].…”

Section: Introductionmentioning

confidence: 99%

“…Modeling and probabilistic framework [1] of robot-grasping tasks have been used to provide insights for future research to grasp planning for robots. Power and dexterity manipulation tasks [2] in grasp planning are important in various tools the robot will grasp in the kitchen environment. A learning approach was used for grasp planning on unseen objects [2] encountered where precision and power for a given object are needed.…”

Section: Introductionmentioning

confidence: 99%

“…Power and dexterity manipulation tasks [2] in grasp planning are important in various tools the robot will grasp in the kitchen environment. A learning approach was used for grasp planning on unseen objects [2] encountered where precision and power for a given object are needed. Even robot learning of grasping unknown objects from human demonstration [3] have been proposed.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

State Classification of Cooking Objects Using a VGG CNN

Mott¹

2019

State Recognition Symposium

View full text Add to dashboard Cite

In machine learning, it is very important for a robot to know the state of an object and recognize particular desired states. This is an image classification problem that can be solved using a convolutional neural network. In this paper, we will discuss the use of a VGG convolutional neural network to recognize those states of cooking objects. We will discuss the uses of activation functions, optimizers, data augmentation, layer additions, and other different versions of architectures. The results of this paper will be used to identify alternatives to the VGG convolutional neural network to improve accuracy.

show abstract

Modeling Grasp Type Improves Learning-Based Grasp Planning

Cited by 35 publications

References 36 publications

Learning better generative models for dexterous, single-view grasping of novel objects

Learning better generative models for dexterous, single-view grasping of novel objects

Automatic precision robot assembly system with microscopic vision and force sensor

State Classification of Cooking Objects Using a VGG CNN

Contact Info

Product

Resources

About