Datasets:
File size: 6,119 Bytes
7a1b06d 224b3ab 7a1b06d 224b3ab 7a1b06d 224b3ab 7a1b06d 224b3ab 7a1b06d 224b3ab 7a1b06d | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 | ---
license: apache-2.0
task_categories:
- robotics
language:
- en
tags:
- LeRobot
configs:
- config_name: default
data_files: data/*/*.parquet
---
This dataset was created using [LeRobot](https://github.com/huggingface/lerobot).
## Dataset Description
This dataset provides the robotic trajectories and observations used in the paper [VITA: Vision-to-Action Flow Matching Policy](https://huggingface.co/papers/2507.13231). VITA introduces a noise-free and conditioning-free policy learning framework that directly maps visual representations to latent actions using flow matching, enabling faster inference for robotic manipulation tasks. The datasets are built on [LeRobot](https://github.com/huggingface/lerobot) Hugging Face formats and optimized into offline `zarr` for faster training.
- **Homepage:** https://ucd-dare.github.io/VITA/
- **Paper:** https://huggingface.co/papers/2507.13231
- **Code:** https://github.com/ucd-dare/VITA
- **License:** apache-2.0
## Dataset Structure
[meta/info.json](meta/info.json):
```json
{
"codebase_version": "v2.1",
"robot_type": null,
"total_episodes": 175,
"total_frames": 26266,
"total_tasks": 1,
"total_videos": 350,
"total_chunks": 1,
"chunks_size": 1000,
"fps": 20,
"splits": {
"train": "0:175"
},
"data_path": "data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet",
"video_path": "videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4",
"features": {
"action": {
"dtype": "float32",
"shape": [
7
],
"names": null
},
"action.delta": {
"dtype": "float32",
"shape": [
7
],
"names": null
},
"action.absolute": {
"dtype": "float32",
"shape": [
7
],
"names": null
},
"observation.state": {
"dtype": "float32",
"shape": [
43
],
"names": null
},
"observation.environment_state": {
"dtype": "float32",
"shape": [
14
],
"names": null
},
"observation.images.agentview_image": {
"dtype": "video",
"shape": [
256,
256,
3
],
"names": [
"height",
"width",
"channel"
],
"info": {
"video.fps": 20.0,
"video.height": 256,
"video.width": 256,
"video.channels": 3,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"has_audio": false
}
},
"observation.images.robot0_eye_in_hand_image": {
"dtype": "video",
"shape": [
256,
256,
3
],
"names": [
"height",
"width",
"channel"
],
"info": {
"video.fps": 20.0,
"video.height": 256,
"video.width": 256,
"video.channels": 3,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"has_audio": false
}
},
"timestamp": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"frame_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"episode_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"index": {
"dtype": "int64",
"shape": [
1
],
"names": null
},
"task_index": {
"dtype": "int64",
"shape": [
1
],
"names": null
}
}
}
```
## Sample Usage
This dataset is designed to be used with the VITA codebase, which extends [LeRobot](https://github.com/huggingface/lerobot). Below are examples for converting datasets to an optimized `zarr` format and training a VITA policy.
First, ensure the VITA repository is cloned and setup, and the `FLARE_DATASETS_DIR` environment variable is set as described in the [VITA GitHub repository](https://github.com/ucd-dare/VITA#%EF%B8%8F-setup).
### Dataset Preprocessing
To list available datasets:
```bash
cd gym-av-aloha/scripts
python convert.py --ls
```
To convert a HuggingFace dataset to an offline `zarr` format (e.g., `av_aloha_sim_hook_package`):
```bash
python convert.py -r iantc104/av_aloha_sim_hook_package
```
### Training a VITA Policy
Once the dataset is converted, you can train a VITA policy using the `flare` module from the VITA codebase:
```bash
python flare/train.py policy=vita task=hook_package session=test
```
You can override default configurations as needed:
```bash
# Example: Use a specific GPU
python flare/train.py policy=vita task=hook_package session=test device=cuda:2
# Example: Change online validation frequency and episodes
python flare/train.py policy=vita task=hook_package session=test \
val.val_online_freq=2000 val.eval_n_episodes=10
# Example: Run an ablation
python flare/train.py policy=vita task=hook_package session=ablate \
policy.vita.decode_flow_latents=False wandb.notes=ablation
```
## Citation
**BibTeX:**
```bibtex
@article{gao2025vita,
title={VITA: Vision-to-Action Flow Matching Policy},
author={Gao, Dechen and Zhao, Boqi and Lee, Andrew and Chuang, Ian and Zhou, Hanchu and Wang, Hang and Zhao, Zhe and Zhang, Junshan and Soltani, Iman},
journal={arXiv preprint arXiv:2507.13231},
year={2025}
}
``` |