metadata
language: en
tags:
- robotics
- grasping
- dual-arm
- SE(3) x SE(3)
- point-cloud
- physics
- dual-arm grasping
license: apache-2.0
datasets:
- dg16m-grasps
size_categories:
- 10K<n<100K
This is the dataset used in the paper, "DAGDiff: Guiding Dual-Arm Grasp Diffusion to Stable and Collision-Free Grasps".
[Project Page] [arXiv] [Github]
Each object in the dataset includes:
- 3D mesh representations of the object
- Signed distance field (SDF) data
- Dual-arm grasp annotations (both positive and negative examples)
- Split files for training and evaluation (e.g.,
train_final.txt,test_final.txt)
Dataset Structure
dagdiff-dataset
├── train_final.txt
├── test_final.txt
|
├── grasps/
│ ├── 554fa306799d623af7248d9dbed7a7b8.h5
│ ├── c2ad96f56ec726d270a43c2d978e502e.h5
│ ├── ....
|
├── meshes/
│ ├── 554fa306799d623af7248d9dbed7a7b8.obj
│ ├── c2ad96f56ec726d270a43c2d978e502e.obj
│ ├── ....
|
└── sdf/
├── 554fa306799d623af7248d9dbed7a7b8.h5
├── c2ad96f56ec726d270a43c2d978e502e.h5
└── ....