# DeformGen Datasets

This repository contains staged release datasets for DeformGen.

## Naming

Paper-facing names are mapped to implementation names as follows:

| Paper name | Code / dataset mode | Meaning |
| --- | --- | --- |
| DG | `yawonly` | yaw-only trajectory warping mode |
| DG* | `txy` | xy translation + yaw trajectory warping mode |
| SMG* | `gridrigid` / `gridrigid_yawonly` | structured grid-rigid state generation data |

The staged directory names intentionally keep implementation names (`txy`, `yawonly`, `gridrigid`) for script compatibility and reproducibility.

## Contents

Each case contains LeRobot-compatible training data and test state lists:

- `rope/train1000_txy`, `rope/train1000_yawonly`, `rope/train1000_gridrigid`, `rope/test200`
- `sloth/train1000_txy`, `sloth/train1000_yawonly`, `sloth/train1000_gridrigid`, `sloth/test200`
- `cloth3/train1000_txy`, `cloth3/train1000_yawonly`, `cloth3/train1000_gridrigid`, `cloth3/test200`

`test200` stores full physics state inputs used by policy evaluation.

## Paper

If you use these artifacts, please cite the DeformGen paper:

- Paper: https://arxiv.org/abs/2606.25939