THULab/demo_three_tasks

Demo Three Tasks - TsFile

This dataset is an Apache TsFile conversion of ethanCSL/demo_three_tasks, which was created using LeRobot.

Modalities: Time-series. The original dataset also contains three RGB camera streams. Those videos are not included here and remain available in the source dataset.

Source Dataset

• Original dataset: ethanCSL/demo_three_tasks
• Pinned source revision: 579c1748a5cb470008eb9022c828af2fe34d276c
• Publisher: ethanCSL
• License: Apache-2.0
• Task category: robotics
• LeRobot codebase version: v3.0
• Robot type: koch_follower
• Split: one train split covering episodes 0:240
• Scale: 240 episodes, 92,668 frames, and 36 frame-data Parquet files; source metadata declares four tasks represented by task indices 0 through 3
• Sampling rate: 30 fps

The repository name says demo_three_tasks, but the authoritative source meta/info.json declares total_tasks: 4, and the frame data contains task_index values 0, 1, 2, and 3. The source meta/tasks.parquet contains these indices but no textual task labels, so this card does not invent task descriptions.

The six action and observation-state elements use the source order: shoulder_pan.pos, shoulder_lift.pos, elbow_flex.pos, wrist_flex.pos, wrist_roll.pos, and gripper.pos.

The source video features are observation.images.front, observation.images.top, and observation.images.wrist. Each is a 480 x 640 RGB AV1 stream at 30 fps. The source repository contains 36 MP4 files per stream, for 108 video files total.

Converted Files

• TsFile: data/demo_three_tasks_train.tsfile
• TsFile size: 2,297,613 bytes
• Rows: 92,668
• Episodes: 240
• Task indices: 4 (0 through 3)
• Table name: demo_three_tasks_train
• Time precision: milliseconds
• Metadata: source meta/ is mirrored, with meta/info.json rewritten to describe the TsFile artifact and conversion mapping.

Schema

Column	Role	Type	Notes
Time	TIME	INT64	round(timestamp * 1000) milliseconds; restarts per episode
episode_index	TAG	INT64	Source episode identifier (0 through 239)
task_index	TAG	INT64	Source task identifier (0 through 3)
frame_index	FIELD	INT64	Source frame index, preserved
sample_index	FIELD	INT64	Renamed from source index
action_0 ... action_5	FIELD	FLOAT	Flattened from source action[6]
observation_state_0 ... observation_state_5	FIELD	FLOAT	Flattened from source observation.state[6]

episode_index and task_index are TAG columns, so individual trajectories or tasks can be selected through the TsFile device dimensions. The file contains 17 columns in total: one TIME column, two TAG columns, and 14 FIELD columns.

Conversion Notes

• The 36 source frame-data Parquet files were merged within the train split into one TsFile.
• Vector columns preserve the complete source column name, replace . with _, and append the element index. Floating-point values remain single-precision FLOAT fields.
• The source timestamp column is omitted because it is redundant with Time / 1000 seconds.
• The source index column is renamed to sample_index; frame_index is retained.
• No source numeric row or other numeric field is intentionally removed.
• Camera videos are neither converted nor uploaded. They remain at: https://huggingface.co/datasets/ethanCSL/demo_three_tasks/tree/main/videos
• Numeric rows preserve episode_index and frame_index, maintaining their alignment with the original source videos.
• meta/info.json records the source and converted paths, pinned revision, actual table name, row and episode counts, TIME/TAG mapping, flattened and renamed fields, omitted video features, and frame/video alignment policy.

Validation

• Pipeline validation confirmed one non-empty TsFile of 2,297,613 bytes.
• Apache TsFile SDK readback returned 92,668 rows, matching the staged Parquet exactly.
• Readback confirmed episode_index and task_index are TAG columns and the remaining 14 measurements are FIELD columns.
• Every staged Time value matches round(source_timestamp * 1000), and no duplicate (episode_index, task_index, Time) key exists.

Read Example

from tsfile import TsFileReader

path = "data/demo_three_tasks_train.tsfile"
table = "demo_three_tasks_train"

with TsFileReader(path) as reader:
    columns = [
        "episode_index",
        "task_index",
        "frame_index",
        "sample_index",
        "action_0",
        "observation_state_0",
    ]
    with reader.query_table(table, columns, batch_size=4096) as result:
        batch = result.read_arrow_batch()
        print(batch)

Citation

The original dataset card does not provide a paper, homepage, or BibTeX citation.