mug_hang_mp_fixed

This repository contains an Apache TsFile conversion of the Hugging Face dataset younghyopark/mug_hang_mp_fixed, a LeRobot-style robotics dataset.

Source Dataset

• Original dataset: younghyopark/mug_hang_mp_fixed
• Author: younghyopark
• License: apache-2.0
• Robot type: DualPanda
• Task: pick something up
• Split: train, episodes 0:10
• Scale: 10 episodes, 11,780 frames, 1 task, 50 fps
• Source data path: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
• Source videos: none (total_videos=0, video_path=null)

Converted Data

• TsFile path: data/mug_hang_mp_fixed.tsfile
• Table name: mug_hang_mp_fixed
• Rows: 11,780
• Columns in staged table: 655
• Time precision: milliseconds
• TAG columns: episode_index, task_index
• FIELD columns: frame_index, sample_index, robot observations, environment state, and action fields
• Mirrored metadata: meta/ is included, with meta/info.json updated to describe the converted TsFile schema

Conversion Notes

Time is synthesized as round(timestamp * 1000) in milliseconds. The source timestamp column is not retained as a separate field because it is equivalent to Time / 1000 seconds. The source index column is renamed to sample_index; frame_index is retained.

Vector and matrix-like source columns are flattened to scalar TsFile fields. The full source column name is preserved by replacing . with _ and appending the flattened element index, for example observation.state becomes observation_state_0 through observation_state_29. Matrix fields such as *.mat are flattened in row-major order.

No videos are uploaded in this converted repository because the source metadata declares no videos.

Minimal Read Example

from tsfile import TsFileReader

path = "data/mug_hang_mp_fixed.tsfile"
reader = TsFileReader(path)
schemas = reader.get_all_table_schemas()
print(schemas.keys())