Datasets:
Add files using upload-large-folder tool
Browse files
README.md
CHANGED
|
@@ -1,3 +1,104 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: cc-by-4.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-4.0
|
| 3 |
+
task_categories:
|
| 4 |
+
- robotics
|
| 5 |
+
tags:
|
| 6 |
+
- lerobot
|
| 7 |
+
- tactile
|
| 8 |
+
- manipulation
|
| 9 |
+
- project-aria
|
| 10 |
+
- egocentric
|
| 11 |
+
size_categories:
|
| 12 |
+
- n<1K
|
| 13 |
+
---
|
| 14 |
+
|
| 15 |
+
# FEEL-Benchmark
|
| 16 |
+
|
| 17 |
+
A multimodal egocentric manipulation dataset combining **Project Aria** smart-glasses recordings (RGB video, eye gaze, hand tracking, IMU) with synchronized **Paxini** tactile-glove force readings. Released in [LeRobot v3.0](https://github.com/huggingface/lerobot) format (GR00T path variant).
|
| 18 |
+
|
| 19 |
+
## Overview
|
| 20 |
+
|
| 21 |
+
- **25 trials**, **650 episodes**, ~1.8 GB
|
| 22 |
+
- **Recording dates**: 2026-01-10, 2026-01-21, 2026-01-28, 2026-04-30, 2026-05-06
|
| 23 |
+
- **Camera**: Project Aria RGB (1408×1408, ~10 fps, H.264)
|
| 24 |
+
- **Tactile**: bilateral force totals (`action.tactile.left.force_total`, `action.tactile.right.force_total`)
|
| 25 |
+
- **Hand pose**: Aria MPS hand tracking (left + right, validity flags, confidence)
|
| 26 |
+
- **Gaze**: Aria MPS general eye gaze (yaw/pitch + depth)
|
| 27 |
+
|
| 28 |
+
## Layout
|
| 29 |
+
|
| 30 |
+
Datasets are grouped by recording session date:
|
| 31 |
+
|
| 32 |
+
```
|
| 33 |
+
FeelAuthors/FEEL-Benchmark/
|
| 34 |
+
├── 20260110/ Jelly_1, Jelly_2, Jelly_3 (3 trials)
|
| 35 |
+
├── 20260121/ Half_coke_2, Half_water_1, Jelly_{5,6,7}, Quarter_coke_0, Quarter_water_2 (7 trials)
|
| 36 |
+
├── 20260128/ Half_water_{3,4}, Sponge_{1,2,3} (5 trials)
|
| 37 |
+
├── 20260430/ Empty_bottle_1, Full_bottle_1 (2 trials)
|
| 38 |
+
└── 20260506/ Chip_{2,3}, Empty_Want_Milk, Empty_wz, P25_bottle, P50_bottle,
|
| 39 |
+
Want_Milk, Wz (8 trials)
|
| 40 |
+
```
|
| 41 |
+
|
| 42 |
+
Each `*_lerobot/` directory follows LeRobot v3.0:
|
| 43 |
+
|
| 44 |
+
```
|
| 45 |
+
<trial>_lerobot/
|
| 46 |
+
├── meta/
|
| 47 |
+
│ ├── info.json # feature schema, fps, episode counts
|
| 48 |
+
│ ├── episodes.jsonl # per-episode metadata
|
| 49 |
+
│ ├── stats.json
|
| 50 |
+
│ └── tasks.jsonl
|
| 51 |
+
├── data/chunk-000/episode_NNNNNN.parquet
|
| 52 |
+
└── videos/chunk-000/main/episode_NNNNNN.mp4
|
| 53 |
+
```
|
| 54 |
+
|
| 55 |
+
## Feature schema
|
| 56 |
+
|
| 57 |
+
Each row in `episode_*.parquet`:
|
| 58 |
+
|
| 59 |
+
| Field | dtype | shape | Notes |
|
| 60 |
+
|------------------------------------|-------------|-------|----------------------------------|
|
| 61 |
+
| `index` | int64 | (1,) | Global frame index |
|
| 62 |
+
| `episode_index` | int64 | (1,) | |
|
| 63 |
+
| `frame_index` | int64 | (1,) | Per-episode frame index |
|
| 64 |
+
| `timestamp` / `timestamp_ns` | float64/int | | Aria device clock |
|
| 65 |
+
| `observation.state` | float32 | (7,) | Right-hand pose state |
|
| 66 |
+
| `action` | float32 | (2,) | Action vector |
|
| 67 |
+
| `status.hands.valid_right` | bool | (1,) | Hand-tracking validity |
|
| 68 |
+
| `action.tactile.left.force_total` | float32 | (1,) | Total force, left glove (N) |
|
| 69 |
+
| `action.tactile.right.force_total` | float32 | (1,) | Total force, right glove (N) |
|
| 70 |
+
| `observation.images.main` | video | 1408×1408×3 | Aria RGB stream |
|
| 71 |
+
|
| 72 |
+
## Path format note
|
| 73 |
+
|
| 74 |
+
Each `meta/info.json` declares `"path_format": "groot"`. The `data_path` and `video_path` templates match standard LeRobot v3.0 (`data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet`), so the dataset is consumable by both stock LeRobot tooling and NVIDIA Isaac GR00T loaders.
|
| 75 |
+
|
| 76 |
+
## Sensors
|
| 77 |
+
|
| 78 |
+
- **Project Aria** (RGB, eye-tracking cameras, SLAM cameras, IMUs) with MPS post-processing for gaze + hand tracking
|
| 79 |
+
- **Paxini tactile glove** for force feedback (totals exported per side; raw per-taxel data not included in this release)
|
| 80 |
+
|
| 81 |
+
## Loading
|
| 82 |
+
|
| 83 |
+
```python
|
| 84 |
+
from datasets import load_dataset
|
| 85 |
+
ds = load_dataset("FeelAuthors/FEEL-Benchmark",
|
| 86 |
+
data_files="20260506/P25_bottle_lerobot/data/chunk-000/*.parquet")
|
| 87 |
+
```
|
| 88 |
+
|
| 89 |
+
Or with LeRobot tooling, point a `LeRobotDataset` at a single trial directory.
|
| 90 |
+
|
| 91 |
+
## License
|
| 92 |
+
|
| 93 |
+
[CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
|
| 94 |
+
|
| 95 |
+
## Citation
|
| 96 |
+
|
| 97 |
+
```bibtex
|
| 98 |
+
@misc{feel_benchmark_2026,
|
| 99 |
+
title = {FEEL-Benchmark: Egocentric Tactile-Visual Manipulation Dataset},
|
| 100 |
+
author = {FeelAuthors},
|
| 101 |
+
year = {2026},
|
| 102 |
+
url = {https://huggingface.co/datasets/FeelAuthors/FEEL-Benchmark}
|
| 103 |
+
}
|
| 104 |
+
```
|