lbm_eval_TurnMugRightsideUp

This dataset was converted from LBM format to LeRobot format.

Dataset Information

• Robot Type: lbm_bimanual_panda
• Total Episodes: 500
• Total Frames: 57276
• FPS: 10
• Cameras: wrist_left_minus, wrist_left_plus, scene_left_0, scene_right_0, wrist_right_plus, wrist_right_minus

Dataset Structure

This dataset follows the LeRobot format:

• data/: Contains episode data in Parquet format
• videos/: Contains video files for each camera
• meta/: Contains metadata files (episodes.jsonl, stats.jsonl, tasks.jsonl, info.json)
• additional_data/ (optional): Contains preserved LBM-specific data (depth, segmentation)

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("lbm_eval_TurnMugRightsideUp")

# Access episode data
episode_data = dataset["train"]

# Get video paths from metadata
import json
with open("meta/episodes.jsonl", "r") as f:
    episodes_meta = [json.loads(line) for line in f]

print(f"First episode videos: {episodes_meta[0]['video_paths']}")

Original Data

This dataset preserves all original LBM data:

• Robot state observations mapped to observation.state.*
• Actions mapped to action.*
• Camera images converted to MP4 videos
• Additional data (depth, segmentation, calibration) preserved in metadata and additional files

Citation

If you use this dataset, please cite the original LBM work and this conversion.