THULab/nevertmr_so101_cup_bluecircle_1task

SO101 Cup Blue Circle 1 Task TsFile

This repository contains an Apache TsFile representation of nevertmr/so101_cup_bluecircle_1task, a LeRobot v3.0 SO-101 dataset for the task pick up the cup near the blue circle and place the cup to the blue circle.

Source and Attribution

• Original dataset: nevertmr/so101_cup_bluecircle_1task
• Original repository creator and uploader: nevertmr
• License: Apache-2.0
• Robot type: so_follower
• LeRobot codebase version: v3.0
• Split: train
• Sampling rate: 30 fps
• Scale: 50 episodes, 19,138 rows, 1 task, 1 source Parquet shard
• Source data layout: data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet
• Source video layout: videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4

The source card identifies no separate paper or citation. Cite the original dataset and nevertmr.

TsFile Contents

• File: data/nevertmr_so101_cup_bluecircle_1task.tsfile
• Table: nevertmr_so101_cup_bluecircle_1task
• Rows: 19,138
• Episode/task devices: 50
• Time precision: milliseconds
• TsFile size: 378,742 bytes

Schema

Time is computed as round(timestamp * 1000) and restarts at zero within each episode.

Columns	TsFile role	Type
Time	TIME	INT64/TIMESTAMP
episode_index, task_index	TAG/device	STRING (source INT64 preserved in metadata)
frame_index, sample_index	FIELD	INT64
action_0 ... action_5	FIELD	FLOAT
observation_state_0 ... observation_state_5	FIELD	FLOAT

The source action[6] and observation.state[6] vectors are flattened while preserving their prefixes. The source index is renamed to sample_index. The source timestamp is omitted after Time synthesis because it is represented by Time / 1000. No numeric rows are discarded.

Storage

• FLOAT/DOUBLE: GORILLA + LZ4
• INT32/INT64: TS_2DIFF + LZ4
• Time: TS_2DIFF + LZ4
• BOOLEAN: RLE + LZ4 when present
• episode_index and task_index: TsFile table-model TAG/device mechanism

Videos and Card Media

Videos are not duplicated in this repository. The original dataset stores frame-aligned streams under:

• observation.images.front
• observation.images.wrist

The source repository currently contains three MP4 container files across these streams. Numeric rows align to the source videos through episode_index and frame_index.

The source dataset card also contains preview_topview.gif, placement_map.png, and placement_grid.png, plus the full collection video. These media files are not included here.

Read Example

from tsfile import TsFileReader

reader = TsFileReader("data/nevertmr_so101_cup_bluecircle_1task.tsfile")
columns = [
    "episode_index",
    "task_index",
    "frame_index",
    "sample_index",
    "action_0",
    "observation_state_0",
]

with reader.query_table(
    "nevertmr_so101_cup_bluecircle_1task",
    columns,
    batch_size=65536,
) as result:
    print(result.read_arrow_batch().to_pandas().head())
reader.close()