Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MultiGripperGrasp: A Dataset for Robotic Grasping from Parallel Jaw Grippers to Dexterous Hands

Luis Felipe Casas, Ninad Khargonkar, Balakrishnan Prabhakaran, Yu Xiang

TL;DR

MultiGripperGrasp introduces a large-scale, cross-gripper grasping dataset comprising 30.4 million grasps across 11 grippers and 345 objects, with grasps labeled by graded fall-off time in the Isaac Sim physics simulation. A key novelty is the alignment of gripper palm poses to enable transfer of grasps between diverse morphologies, significantly increasing usable grasps for each gripper. Grasp generation combines GraspIt! candidate proposals with Isaac Sim filtering for ranking, producing a graded quality measure beyond binary success. Experiments on grasp ranking and cross-gripper transfer demonstrate the value of grasp transfer for augmenting underperforming grippers and highlight controller-related limitations, setting the stage for generalized grasp planning and cross-domain skill transfer in robotics.

Abstract

We introduce a large-scale dataset named MultiGripperGrasp for robotic grasping. Our dataset contains 30.4M grasps from 11 grippers for 345 objects. These grippers range from two-finger grippers to five-finger grippers, including a human hand. All grasps in the dataset are verified in the robot simulator Isaac Sim to classify them as successful and unsuccessful grasps. Additionally, the object fall-off time for each grasp is recorded as a grasp quality measurement. Furthermore, the grippers in our dataset are aligned according to the orientation and position of their palms, allowing us to transfer grasps from one gripper to another. The grasp transfer significantly increases the number of successful grasps for each gripper in the dataset. Our dataset is useful to study generalized grasp planning and grasp transfer across different grippers. Data, code and videos for the project are available at https://irvlutd.github.io/MultiGripperGrasp

MultiGripperGrasp: A Dataset for Robotic Grasping from Parallel Jaw Grippers to Dexterous Hands

TL;DR

MultiGripperGrasp introduces a large-scale, cross-gripper grasping dataset comprising 30.4 million grasps across 11 grippers and 345 objects, with grasps labeled by graded fall-off time in the Isaac Sim physics simulation. A key novelty is the alignment of gripper palm poses to enable transfer of grasps between diverse morphologies, significantly increasing usable grasps for each gripper. Grasp generation combines GraspIt! candidate proposals with Isaac Sim filtering for ranking, producing a graded quality measure beyond binary success. Experiments on grasp ranking and cross-gripper transfer demonstrate the value of grasp transfer for augmenting underperforming grippers and highlight controller-related limitations, setting the stage for generalized grasp planning and cross-domain skill transfer in robotics.

Abstract

We introduce a large-scale dataset named MultiGripperGrasp for robotic grasping. Our dataset contains 30.4M grasps from 11 grippers for 345 objects. These grippers range from two-finger grippers to five-finger grippers, including a human hand. All grasps in the dataset are verified in the robot simulator Isaac Sim to classify them as successful and unsuccessful grasps. Additionally, the object fall-off time for each grasp is recorded as a grasp quality measurement. Furthermore, the grippers in our dataset are aligned according to the orientation and position of their palms, allowing us to transfer grasps from one gripper to another. The grasp transfer significantly increases the number of successful grasps for each gripper in the dataset. Our dataset is useful to study generalized grasp planning and grasp transfer across different grippers. Data, code and videos for the project are available at https://irvlutd.github.io/MultiGripperGrasp
Paper Structure (15 sections, 8 figures, 3 tables)

This paper contains 15 sections, 8 figures, 3 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. Grasping Datasets
  4. Grasp Synthesis Approaches
  5. Learning-based Grasping
  6. METHOD: Dataset Generation
  7. Objects and Grippers
  8. Grasp Generation
  9. Grasp Labeling
  10. Controller Design
  11. EXPERIMENTS
  12. Grasp Ranking
  13. Grasp Transfer
  14. LIMITATIONS
  15. CONCLUSION

Figures (8)

  • Figure 1: Dataset Visualization: Our dataset contains grasps from 11 different grippers on a diverse object set.
  • Figure 2: Our dataset creation pipeline: Given a pair of an object and a gripper, GraspIt! is used to generate candidate grasps that are then filtered using Isaac Sim to produce ranked grasps.
  • Figure 3: A sample set of objects used for grasp generation.
  • Figure 4: Aligned grippers in our dataset with varying dexterity.
  • Figure 5: Illustration of the grasp ranking process with Isaac Sim.
  • ...and 3 more figures