Factory-Intelligence/NIRANJAN_one_wire_merged_20260415_dagger_iter1
收藏Hugging Face2026-04-16 更新2026-04-26 收录
下载链接:
https://hf-mirror.com/datasets/Factory-Intelligence/NIRANJAN_one_wire_merged_20260415_dagger_iter1
下载链接
链接失效反馈官方服务:
资源简介:
---
license: apache-2.0
task_categories:
- robotics
tags:
- LeRobot
- dagger
- hg-dagger
- yam
- act-policy-rollouts
- wire-manipulation
configs:
- config_name: default
data_files: data/*/*.parquet
---
This dataset was created using [LeRobot](https://github.com/huggingface/lerobot).
## Dataset Description
## HG-DAgger iteration 1 for `Factory-Intelligence/NIRANJAN_one_wire_merged_20260415`
Human-Gated DAgger (HG-DAgger) rollouts for the task **"Pick up only one wire from the bin"** on a **Single YAM follower (right arm)**.
The base ACT policy was executed on real hardware while a human expert could take
over the leader arm at any point by pressing **T**; on release, policy control
resumed with a 10-frame blend for smoothness. Every frame is labeled with
`is_intervention` so a fine-tune can upweight corrected frames.
### Pipeline
- **Base dataset:** [Factory-Intelligence/NIRANJAN_one_wire_merged_20260415](https://huggingface.co/datasets/Factory-Intelligence/NIRANJAN_one_wire_merged_20260415)
- **Base policy (used for rollouts):** `outputs/train/act_one_wire_merged/checkpoints/265000/pretrained_model` (ACT, chunk_size=100, 7-DOF action)
- **Collection script:** `python -m lerobot.dagger.collect ...` (from [this fork](https://github.com/3d-FI/lerobot))
### Extra feature vs. base
| Feature | Shape | dtype | Semantics |
|---------|-------|-------|-----------|
| `is_intervention` | (1,) | float32 | `1.0` if the expert was driving that frame, `0.0` if the base policy was |
### Session stats
- **Episodes:** 20
- **Total frames:** 17,452 (9.7 minutes at 30 fps)
- **Intervention frames:** 3,587 (20.6% of total)
#### Per-episode breakdown
| Episode | Frames | Intervention Frames | Intervention % |
|---------|--------|---------------------|----------------|
| 0 | 513 | 0 | 0.0% |
| 1 | 1146 | 56 | 4.9% |
| 2 | 496 | 0 | 0.0% |
| 3 | 691 | 0 | 0.0% |
| 4 | 787 | 53 | 6.7% |
| 5 | 1211 | 461 | 38.1% |
| 6 | 1230 | 198 | 16.1% |
| 7 | 975 | 421 | 43.2% |
| 8 | 530 | 0 | 0.0% |
| 9 | 917 | 357 | 38.9% |
| 10 | 1173 | 0 | 0.0% |
| 11 | 943 | 0 | 0.0% |
| 12 | 1127 | 311 | 27.6% |
| 13 | 928 | 355 | 38.3% |
| 14 | 901 | 330 | 36.6% |
| 15 | 505 | 0 | 0.0% |
| 16 | 1065 | 256 | 24.0% |
| 17 | 567 | 0 | 0.0% |
| 18 | 905 | 458 | 50.6% |
| 19 | 842 | 331 | 39.3% |
### How to use with fine-tuning
```bash
# 1. Merge base + all DAgger iters into one training dataset (auto-detects iter{N})
python -m lerobot.dagger.merge \
--base_repo_id=Factory-Intelligence/NIRANJAN_one_wire_merged_20260415 \
--output_repo_id=Factory-Intelligence/NIRANJAN_one_wire_merged_20260415_merged_dagger_v1
# 2. Fine-tune the base policy, with 3x loss weight on is_intervention=1 frames
python -m lerobot.dagger.train \
--policy.path=<path to previous pretrained_model> \
--merged_repo_id=Factory-Intelligence/NIRANJAN_one_wire_merged_20260415_merged_dagger_v1 \
--output_dir=outputs/dagger/iter1 \
--steps=20000 --intervention_weight=3.0
```
### Notes
- Recorded at 30 fps with streaming h264_nvenc encoding.
- Cameras: `cam_right` (wrist), `cam_global` (overhead). Both 640×480 @ 30fps MJPG.
- This dataset was **interrupted mid-session** by a transient USB disconnect of the
global camera; the 20 episodes saved before the crash are complete and valid.
- **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",
"robot_type": "yam_follower",
"total_episodes": 20,
"total_frames": 17452,
"total_tasks": 1,
"chunks_size": 1000,
"data_files_size_in_mb": 100,
"video_files_size_in_mb": 200,
"fps": 30,
"splits": {
"train": "0:20"
},
"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",
"features": {
"action": {
"dtype": "float32",
"names": [
"joint_0.pos",
"joint_1.pos",
"joint_2.pos",
"joint_3.pos",
"joint_4.pos",
"joint_5.pos",
"gripper.pos"
],
"shape": [
7
]
},
"observation.state": {
"dtype": "float32",
"names": [
"joint_0.pos",
"joint_1.pos",
"joint_2.pos",
"joint_3.pos",
"joint_4.pos",
"joint_5.pos",
"gripper.pos"
],
"shape": [
7
]
},
"observation.images.cam_right": {
"dtype": "video",
"shape": [
480,
640,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 480,
"video.width": 640,
"video.codec": "h264",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"video.fps": 30,
"video.channels": 3,
"has_audio": false
}
},
"observation.images.cam_global": {
"dtype": "video",
"shape": [
480,
640,
3
],
"names": [
"height",
"width",
"channels"
],
"info": {
"video.height": 480,
"video.width": 640,
"video.codec": "h264",
"video.pix_fmt": "yuv420p",
"video.is_depth_map": false,
"video.fps": 30,
"video.channels": 3,
"has_audio": false
}
},
"is_intervention": {
"dtype": "float32",
"shape": [
1
],
"names": null
},
"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
}
}
}
```
## Citation
**BibTeX:**
```bibtex
[More Information Needed]
```
提供机构:
Factory-Intelligence



