mug_hang_mp_500 (TsFile)

This repository contains a TsFile conversion of the Hugging Face dataset younghyopark/mug_hang_mp_500, a LeRobot-format robotics dataset created with LeRobot.

Modalities: Time-series.

Source Dataset

• Original dataset: younghyopark/mug_hang_mp_500
• License: apache-2.0
• Robot type: DualPanda
• Codebase version: v2.1
• Episodes: 500
• Frames: 591,679
• Tasks: 1
• FPS: 50
• Original data path pattern: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet
• Original video path: None; total videos: 0

Converted Files

• Converted rows: 591,679

• TsFile shards: 6

• Total TsFile size: 939,574,292 bytes

• Table name: mug_hang_mp_500

• Converted data path pattern recorded in meta/info.json: data/mug_hang_mp_500_*.tsfile

• data/mug_hang_mp_500_1.tsfile (158,338,984 bytes)

• data/mug_hang_mp_500_2.tsfile (158,768,860 bytes)

• data/mug_hang_mp_500_3.tsfile (159,295,248 bytes)

• data/mug_hang_mp_500_4.tsfile (159,249,339 bytes)

• data/mug_hang_mp_500_5.tsfile (158,149,388 bytes)

• data/mug_hang_mp_500_6.tsfile (145,772,473 bytes)

TsFile Schema Design

• Time is the TsFile time column in millisecond precision.
• episode_index and task_index are TAG columns. They identify the LeRobot episode and task.
• FIELD columns: 652 scalar measurements after flattening vector and matrix features.
• The original timestamp field is dropped because Time = round(timestamp * 1000) and therefore represents the same value in milliseconds.
• The original index column is renamed to sample_index; frame_index is retained.
• Source feature names containing . are converted to _ in TsFile field names.

Flattening examples

Source feature	Converted field range	Count
observation.joint	observation_joint_0 ... observation_joint_15	16
observation.state	observation_state_0 ... observation_state_29	30
observation.environment_state	observation_environment_state_0 ... observation_environment_state_13	14
observation.ee_pose	observation_ee_pose_0 ... observation_ee_pose_15	16
action.ctrl	action_ctrl_0 ... action_ctrl_15	16
action.joint	action_joint_0 ... action_joint_15	16
action.ee_pose	action_ee_pose_0 ... action_ee_pose_15	16

Array2D-style matrix features are flattened in row-major order. For example, a 3x3 matrix becomes 9 scalar FLOAT fields with suffixes _0 through _8.

Metadata

The source meta/ directory is mirrored under meta/. The file meta/info.json has been rewritten so that:

• data_path points to data/mug_hang_mp_500_*.tsfile instead of the original Parquet pattern.
• features describes the converted TsFile scalar schema and roles (TIME, TAG, FIELD).
• tsfile_conversion records the source dataset, time mapping, row count, flattened features, renamed fields, and dropped fields.

Reading

from tsfile import TsFileReader

reader = TsFileReader("data/mug_hang_mp_500_1.tsfile")
# Inspect available table schemas or query table data with the Apache TsFile Python SDK.
print(reader.get_all_table_schemas())
reader.close()

Conversion Notes

Conversion notes:

• LeRobot-style robot dataset converted to TsFile via the generic lerobot converter.
• Source metadata declares robot_type=DualPanda, 500 episodes, 591,679 frames, 1 task, 50 fps, and Apache-2.0 license.
• Numeric frame data is stored as one logical TsFile table under data/. Because this dataset is wide, local conversion uses bounded Parquet row groups and may emit data/mug_hang_mp_500_*.tsfile shards.
• Source columns episode_index and task_index are TAG columns; no synthetic episode_id or task_id columns are created. Query a single episode with WHERE episode_index=N.
• Vector columns preserve the full source column name when flattened: . is replaced with _ and the element index is appended. Values are stored as single-precision FLOAT fields.
• Time is synthesized as round(timestamp * 1000) milliseconds. The source timestamp column is not retained as a separate field because it equals Time / 1000 seconds; frame_index is kept.
• Source metadata has total_videos=0 and video_path=null; no videos are included in this converted repository.
• meta/ is mirrored from the source except that meta/info.json describes the converted TsFile schema and records the original video path separately.