Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

QDGset: A Large Scale Grasping Dataset Generated with Quality-Diversity

Johann Huber, François Hélénon, Mathilde Kappel, Ignacio de Loyola Páez-Ubieta, Santiago T. Puente, Pablo Gil, Faïz Ben Amar, Stéphane Doncieux

TL;DR

This work proposes a data augmentation method that combines the transformation of object meshes with transfer learning from previous grasping repertoires and uses this approach to generate QDGset, a dataset of 6 DoF grasp poses that contains about 3.5 and 4.5 times more grasps and objects than the previous state-of-the-art.

Abstract

Recent advances in AI have led to significant results in robotic learning, but skills like grasping remain partially solved. Many recent works exploit synthetic grasping datasets to learn to grasp unknown objects. However, those datasets were generated using simple grasp sampling methods using priors. Recently, Quality-Diversity (QD) algorithms have been proven to make grasp sampling significantly more efficient. In this work, we extend QDG-6DoF, a QD framework for generating object-centric grasps, to scale up the production of synthetic grasping datasets. We propose a data augmentation method that combines the transformation of object meshes with transfer learning from previous grasping repertoires. The conducted experiments show that this approach reduces the number of required evaluations per discovered robust grasp by up to 20%. We used this approach to generate QDGset, a dataset of 6DoF grasp poses that contains about 3.5 and 4.5 times more grasps and objects, respectively, than the previous state-of-the-art. Our method allows anyone to easily generate data, eventually contributing to a large-scale collaborative dataset of synthetic grasps.

QDGset: A Large Scale Grasping Dataset Generated with Quality-Diversity

TL;DR

This work proposes a data augmentation method that combines the transformation of object meshes with transfer learning from previous grasping repertoires and uses this approach to generate QDGset, a dataset of 6 DoF grasp poses that contains about 3.5 and 4.5 times more grasps and objects than the previous state-of-the-art.

Abstract

Recent advances in AI have led to significant results in robotic learning, but skills like grasping remain partially solved. Many recent works exploit synthetic grasping datasets to learn to grasp unknown objects. However, those datasets were generated using simple grasp sampling methods using priors. Recently, Quality-Diversity (QD) algorithms have been proven to make grasp sampling significantly more efficient. In this work, we extend QDG-6DoF, a QD framework for generating object-centric grasps, to scale up the production of synthetic grasping datasets. We propose a data augmentation method that combines the transformation of object meshes with transfer learning from previous grasping repertoires. The conducted experiments show that this approach reduces the number of required evaluations per discovered robust grasp by up to 20%. We used this approach to generate QDGset, a dataset of 6DoF grasp poses that contains about 3.5 and 4.5 times more grasps and objects, respectively, than the previous state-of-the-art. Our method allows anyone to easily generate data, eventually contributing to a large-scale collaborative dataset of synthetic grasps.
Paper Structure (6 sections, 2 equations, 9 figures, 1 table)

This paper contains 6 sections, 2 equations, 9 figures, 1 table.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORKS
  3. METHOD
  4. EXPERIMENTS
  5. RESULTS AND DISCUSSION
  6. CONCLUSIONS

Figures (9)

  • Figure 1: QDGset contains 62M 6DoF grasps on about 40k simulated objects. Each grasp is labelled with a probability to transfer in the real world huber2023domainrandomization.
  • Figure 2: Data augmentation principle. A gripper and an object model are provided to QDG6DoF huber2024speeding to generate grasp poses. The object model is then perturbed to generate new objects leveraged in new QDG6DoF runs. Each of those runs are bootstrapped with previously found grasps.
  • Figure 3: QDGset objects. (Left) Number of objects per dataset; (center) ratio of augmented objects; (right) object categories.
  • Figure 4: Object sizes. (Left) QDGset object size distribution before ShapeNet rescale; (right) shapenet rescale based on YCB sizes.
  • Figure 5: Number of successful grasps per object. Most of the distributions lie between 1000 and 5000 grasps per objects. Those distributions reflect the complexity of the subset of objects.
  • ...and 4 more figures