DatasetsT1 / README.md
KillzoneCCO's picture
Upload README.md with huggingface_hub
4845dfc verified
|
Raw History Blame Contribute Delete
1.96 kB
---
tags:
- robotics
- lerobot
- booster-t1
- recovery
task_categories:
- robotics
---
# Booster T1 Task Recovery Dataset (LeRobot v3.0)
## 🎥 Video Stream Preview (Head RGB Camera @ 8.0 FPS)
<video controls width="720" src="https://huggingface.co/datasets/KillzoneCCO/Booster-T1-Recovery/resolve/main/videos/observation.images.head_camera/chunk-000/file-000.mp4"></video>
> ℹ️ **Прямой просмотр видеопотока:** Видеофайл эпизода 0: [file-000.mp4](https://huggingface.co/datasets/KillzoneCCO/Booster-T1-Recovery/resolve/main/videos/observation.images.head_camera/chunk-000/file-000.mp4).
> Все 288 видео доступны в каталоге `videos/observation.images.head_camera/chunk-000/` (от `file-000.mp4` до `file-287.mp4`).
---
## 📊 Dataset Specifications
- **Codebase Version:** LeRobot v3.0 (`meta/info.json`)
- **Robot:** Booster T1 (`Robot-bc43b7`)
- **Task:** Transferring plates from the table to the container (Automated pick-and-place with teleoperation recovery)
- **Total Episodes:** 288 continuous manipulation segments (30 parent teleoperation recoveries)
- **Total Frames:** 8,785 frames @ 8.0 FPS
- **Action Shape:** `[20]` DOF (float32, radians)
- **State Shape:** `[37]` DOF (float32, joint positions and base pose)
- **Primary Camera:** `observation.images.head_camera` (1280x720 H.264 @ 8.0 FPS)
- **QC Quality:** 100% Passed (0 NaNs, 0 Infs, 0 Timestamp Gaps)
---
## 🚀 How to Load in Python (Official LeRobot v3 Loader)
```python
from lerobot.datasets.lerobot_dataset import LeRobotDataset
# Загрузка полного датасета с Hugging Face:
dataset = LeRobotDataset("KillzoneCCO/Booster-T1-Recovery")
print(f"Total episodes: {dataset.meta.total_episodes}")
print(f"Total frames: {len(dataset)}")
sample = dataset[0]
print(f"Action shape: {sample['action'].shape}")
print(f"State shape: {sample['observation.state'].shape}")
```