---
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, "<a href="https://arxiv.org/pdf/2509.21145">DAGDiff: Guiding Dual-Arm Grasp Diffusion to Stable and Collision-Free Grasps</a>". 

<a href="https://dag-diff.github.io/dagdiff/">[Project Page]</a> <span> <a href="https://arxiv.org/abs/2509.21145">[arXiv]</a> <span> <a href="https://github.com/DAG-Diff/dual-arm-grasp-diffusion">[Github]</a>

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
    └── ....
```