leapshared's picture
Redox 서브태스크 데이터에 경계 멈춤을 넣은 사본 (검전기 멈춤 추가본과 같은 규칙)
fc55502 verified
|
Raw History Blame Contribute Delete
9.5 kB
---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
configs:
- config_name: default
data_files: data/*/*.parquet
---
This dataset was created using [LeRobot](https://github.com/huggingface/lerobot).
<a class="flex" href="https://huggingface.co/spaces/lerobot/visualize_dataset?path=leapshared/Redox_131epi_subtask_pause_added">
<img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl.svg"/>
<img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface/badges/resolve/main/visualize-this-dataset-xl-dark.svg"/>
</a>
## 이 데이터셋에 대해
`leapshared/Redox_131epi_subtask`에 멈춤을 넣은 사본이다(2026-10-01). 검전기 멈춤 추가본
(`leapshared/Electroscope_113epi_subtask_pause_added`)과 같은 규칙이다. 시연자가 원래 멈춘 프레임을 90번(3초) 복사해 넣었다.
복사본의 앞 40개는 지금 구간 문장, 뒤 50개는 다음 구간 문장이다. 복사하는 경계는 멈춤 프레임 바로 뒤로 옮겼다.
문장을 하나 받으면 그 동작만 하고 멈추고, 다음 문장을 받아야 움직이게 하려는 것이다.
| 경계 | 처리 | 멈춤 프레임 |
|---|---|---|
| ① 1→2 | 복사 | 판을 쥔 직후(오른그리퍼가 다 닫힌 뒤 가장 느린 프레임). 들어 올린 뒤에 멈춘 시연은 없다 |
| ② 2→3 | 복사 | 판이 파란 용액 위에 도착하고, 왼손이 가리키러 가기 전 |
| ③ 3→4 | 복사 | 왼손이 판을 가리킨 채 멈춘 끝(거두기 전) |
| 4→5 | 그대로 | 내려놓으면 곧 담근 채 기다림이라 이미 멈춘다 |
| ⑤ 5→6 (두 번) | 복사 | 담근 채 기다림의 끝 |
| ⑥ 6→4, 6→7 | 복사 | 올려서 가리킨 채 멈춘 끝 |
| 7→8 | 그대로 | 검전기의 6→7처럼 내려놓은 뒤 홈으로 간다 |
그래서 문장마다 배우는 것은 이렇다.
- 1번: 판을 쥐고 멈춘다.
- 2번: 들어 올려 용액 위로 옮기고 멈춘다.
- 3번: 왼손으로 가리킨 채 멈춘다.
- 4번: 왼손을 거두고 판을 담근다.
- 5번: 담근 채 멈춘다.
- 6번: 올려서 가리킨 채 멈춘다.
- 7번: 왼손을 거두고 왼쪽 비커에 넣은 뒤 홈으로 간다.
③⑥을 왼손을 거둔 뒤가 아니라 가리킨 채에 두었다. 2·3·6번이 끝나는 자세(판은 용액 위, 왼손은 옆)가 손끝 2 cm 안으로 같아서다.
거둔 뒤에 복사하면 그 자세와 3번 문장에 "가리키러 간다"와 "멈춘다"가 함께 생긴다(검전기에서 되풀이 구간 끝에 복사하지 않은 것과 같은 이유).
- 에피소드 131개(번호는 원본과 같다). 프레임 154,675 → 218,485(+41%), 복사 709곳.
- 학습 목록은 `train_episodes.json`(110개)이다. 원본 목록 121개에서 멈춤 프레임의 팔 속도가
20°/s를 넘는 11개(11, 21, 25, 46, 51, 55, 59, 60, 71, 76, 98)를 뺐다. 대부분 왼손이 가리키러 가기 시작할 때 오른손이 아직 움직이던
에피소드다. 검증 에피소드는 7, 27, 47, 67, 87, 107, 127이다(47은 원본이 통째로 1번 라벨이라 복사하지 않았다).
- 에피소드마다 경계, 멈춤 프레임, 속도는 `pause_plan.json`에 있다. 만든 도구는 H200 작업 폴더의 `scripts/redox_stops.py`(멈춤 프레임.
손끝 위치는 openarm_bimanual.xml 순기구학), `scripts/insert_stop_frames.py`(만들기), `scripts/check_stop_frames.py`(대조)다.
- 영상은 원본과 같은 설정(libsvtav1, crf 30, g 2, preset 12)으로 다시 인코딩했다. 이 서버의 SVT-AV1 3.0은 preset 12를 10으로 돈다.
짝 프레임과의 평균 차이는 1.4~1.7(0~255)이고, ±3프레임 어긋난 것과는 4~9다.
- `meta/stats.json`은 LeRobot이 에피소드 통계를 합친 것이다. 분위수는 학습 전에 전체 프레임으로 다시 구한다(H200 체인의 preflight).
## Dataset Description
- **Homepage:** [More Information Needed]
- **Paper:** [More Information Needed]
- **License:** apache-2.0
## Dataset Structure
[meta/info.json](meta/info.json):
```json
{
"codebase_version": "v3.0",
"fps": 30,
"features": {
"action": {
"dtype": "float32",
"names": [
"left_joint_1.pos",
"left_joint_2.pos",
"left_joint_3.pos",
"left_joint_4.pos",
"left_joint_5.pos",
"left_joint_6.pos",
"left_joint_7.pos",
"left_gripper.pos",
"right_joint_1.pos",
"right_joint_2.pos",
"right_joint_3.pos",
"right_joint_4.pos",
"right_joint_5.pos",
"right_joint_6.pos",
"right_joint_7.pos",
"right_gripper.pos"
],
"shape": [
16
]
},
"observation.state": {
"dtype": "float32",
"names": [
"left_joint_1.pos",
"left_joint_2.pos",
"left_joint_3.pos",
"left_joint_4.pos",
"left_joint_5.pos",
"left_joint_6.pos",
"left_joint_7.pos",
"left_gripper.pos",
"right_joint_1.pos",
"right_joint_2.pos",
"right_joint_3.pos",
"right_joint_4.pos",
"right_joint_5.pos",
"right_joint_6.pos",
"right_joint_7.pos",
"right_gripper.pos"
],
"shape": [
16
]
},
"observation.images.follower_d455f": {
"dtype": "video",
"shape": [
480,
640,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 480,
"video.width": 640,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.fps": 30,
"video.channels": 3,
"has_audio": false,
"video.g": 2,
"video.crf": 30,
"video.preset": 12,
"video.fast_decode": 0,
"video.video_backend": "pyav",
"video.extra_options": {},
"is_depth_map": false
}
},
"observation.images.left_wrist": {
"dtype": "video",
"shape": [
480,
640,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 480,
"video.width": 640,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.fps": 30,
"video.channels": 3,
"has_audio": false,
"video.g": 2,
"video.crf": 30,
"video.preset": 12,
"video.fast_decode": 0,
"video.video_backend": "pyav",
"video.extra_options": {},
"is_depth_map": false
}
},
"observation.images.right_wrist": {
"dtype": "video",
"shape": [
480,
640,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 480,
"video.width": 640,
"video.codec": "av1",
"video.pix_fmt": "yuv420p",
"video.fps": 30,
"video.channels": 3,
"has_audio": false,
"video.g": 2,
"video.crf": 30,
"video.preset": 12,
"video.fast_decode": 0,
"video.video_backend": "pyav",
"video.extra_options": {},
"is_depth_map": 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
}
},
"total_episodes": 131,
"total_frames": 218485,
"total_tasks": 8,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"data_path": "data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet",
"video_path": "videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4",
"robot_type": "bi_openarm_follower",
"splits": {
"train": "0:131"
}
}
```
## Citation
**BibTeX:**
```bibtex
[More Information Needed]
```