livsynrobotics/LivUMI-Ego-Sample

livego-lite-hand-tracking

LeRobot Dataset v3.0 for dual-hand palm pose tracking with synchronized videos.

Overview

• Robot type: livego-lite
• Codebase version: v3.0
• Episodes: 8
• Frames: 1732
• FPS: 11.2356
• Task count: 1
• Train split: 0:8

This dataset contains:

• Low-dimensional hand state in data/chunk-000/file-000.parquet
• Two video streams per episode:
  • observation.images.head (original video)
  • observation.images.head_processed (processed video)
• Metadata in meta/ (info.json, stats.json, tasks.parquet, episodes/*.parquet)

Directory Structure

livego-lite-hand-tracking/
├── data/
│   └── chunk-000/
│       └── file-000.parquet
├── meta/
│   ├── info.json
│   ├── stats.json
│   ├── tasks.parquet
│   └── episodes/
│       └── chunk-000/
│           └── file-000.parquet
└── videos/
    ├── observation.images.head/
    │   └── chunk-000/
    │       ├── file-000.mp4
    │       └── ...
    └── observation.images.head_processed/
        └── chunk-000/
            ├── file-000.mp4
            └── ...

Data Schema

Main frame table (data/chunk-000/file-000.parquet) columns:

• observation.images.head: relative path to original video shard
• observation.images.head_processed: relative path to processed video shard
• observation.state: 16-dim float vector
• timestamp: timestamp_bj_ms
• frame_index: point_index
• episode_index: episode id in dataset
• index: global frame index
• task_index: task id

observation.state order:

1. left_hand.valid (0/1)
2. left_hand.position.x
3. left_hand.position.y
4. left_hand.position.z
5. left_hand.rotation.x
6. left_hand.rotation.y
7. left_hand.rotation.z
8. left_hand.rotation.w
9. right_hand.valid (0/1)
10. right_hand.position.x
11. right_hand.position.y
12. right_hand.position.z
13. right_hand.rotation.x
14. right_hand.rotation.y
15. right_hand.rotation.z
16. right_hand.rotation.w

Value Conventions

• hand_id=0 in source CSV => left hand
• hand_id=1 in source CSV => right hand
• palm_x, palm_y, palm_z_mm are converted from millimeters to meters
• roll_deg, pitch_deg, yaw_deg are converted:
  • degrees -> radians
  • euler -> quaternion (x, y, z, w)

Quick Start (Python)

from pathlib import Path
import pandas as pd

root = Path("livego-lite-hand-tracking")
df = pd.read_parquet(root / "data/chunk-000/file-000.parquet")
tasks = pd.read_parquet(root / "meta/tasks.parquet")
episodes = pd.read_parquet(root / "meta/episodes/chunk-000/file-000.parquet")

print(df.head(2))
print(tasks)
print(episodes[["episode_index", "length", "dataset_from_index", "dataset_to_index"]])

Notes

• The dataset currently contains only a train split (0:8).
• Global statistics for normalization are provided in meta/stats.json.