THULab/Jiangeng_so100_413

SO100 413 TsFile

This dataset is an Apache TsFile representation of Jiangeng/so100_413, a LeRobot v2.1 SO-100 dataset for the task "Pick the cube."

The numeric trajectories, episode and task identifiers, frame timing, and source metadata are included. Camera videos remain in the original Hugging Face repository.

Source Dataset

• Source: Jiangeng/so100_413
• Author and publisher: Jiangeng, the repository owner and sole listed contributor
• License: Apache-2.0
• Robot: SO-100
• LeRobot codebase version: v2.1
• Task: Pick the cube
• Split: train, episodes 0 through 49
• Scale: 50 episodes, 17,851 frames, 1 task, 30 fps
• Source layout: 50 Parquet episode shards in data/chunk-000/
• Source numeric size: 1,114,470 bytes
• TsFile table: jiangeng_so100_413
• TsFile rows: 17,851

The original card does not identify a separate personal author name, paper, homepage, or complete citation. Attribution therefore follows the facts published by the source repository.

Schema

Column	TsFile role	Type	Source
Time	TIME	millisecond timestamp	round(timestamp * 1000)
episode_index	TAG	string device segment	source scalar
task_index	TAG	string device segment	source scalar
frame_index	FIELD	INT64	source scalar
sample_index	FIELD	INT64	source index
action_0 ... action_5	FIELD	FLOAT	flattened action[6]
observation_state_0 ... observation_state_5	FIELD	FLOAT	flattened observation.state[6]

The six action and state dimensions are, in source order: left_shoulder_pan, left_shoulder_lift, left_elbow_flex, left_wrist_flex, left_wrist_roll, and left_gripper.

Time And Episode Mapping

Time is round(timestamp * 1000) in milliseconds. It restarts at zero for each episode and remains monotonic within each (episode_index, task_index) device. The original timestamp field is not repeated because it equals Time / 1000 seconds. frame_index is retained, and source index is renamed to sample_index.

All 50 source episode shards are merged into one table-model TsFile. No numeric rows, episodes, tasks, action dimensions, or state dimensions are omitted.

Videos

The source contains 100 AV1 MP4 files at 30 fps, split across two camera streams with 50 episode files each:

• videos/chunk-000/observation.images.top/
• videos/chunk-000/observation.images.wrist_left/

Videos are not included here. Use episode_index and frame_index to align numeric rows with the corresponding source video frames.

Usage

from tsfile import ColumnCategory, TsFileReader

path = "data/jiangeng_so100_413.tsfile"
reader = TsFileReader(path)
table_name = "jiangeng_so100_413"
schema = reader.get_all_table_schemas()[table_name]
columns = [
    column.get_column_name()
    for column in schema.get_columns()
    if column.get_category() in (ColumnCategory.FIELD, ColumnCategory.TAG)
]

with reader.query_table(table_name, columns, batch_size=65536) as result:
    first_batch = result.read_arrow_batch()

Citation

The source repository does not provide a complete BibTeX citation. Please cite the original dataset URL and credit Jiangeng when using this data.