azithromycin/jclinton1_wedgit_stack_single_dual_cam

Wedgit Stack Single Dual Cam TsFile

This Apache TsFile dataset is derived from jclinton1/wedgit_stack_single_dual_cam, a LeRobot v2.1 so100_with_koch dataset for stacking a yellow Wedgit block on a blue block.

Source dataset

• Repository owner and uploader: Joe Clinton (jclinton1)
• License: Apache-2.0
• Paper and formal citation: not provided by the source repository
• Split: train; sampling frequency: 30 fps
• Task 0: Stack the yellow wedgit block on the blue one
• Scale: 101 episodes, 31,354 frame rows, one task, 101 Parquet shards
• Source data path: data/chunk-{episode_chunk:03d}/episode_{episode_index:06d}.parquet

TsFile contents

• File: data/jclinton1_wedgit_stack_single_dual_cam.tsfile
• Table: jclinton1_wedgit_stack_single_dual_cam
• Rows: 31,354; episode devices: 101
• Time precision: milliseconds
• Source Parquet size: 1,829,132 bytes
• TsFile size: 605,066 bytes
• TsFile/source size ratio: 0.330794

The source meta/ files are retained. meta/info.json points data_path to the TsFile and records the source video location and frame-alignment policy.

Schema

Column or group	TsFile role	Type	Description
Time	TIME	INT64	round(timestamp * 1000) ms; restarts at zero per episode
episode_index	TAG	STRING	Original episode index stored through the device/TAG mechanism
task_index	TAG	STRING	Original task index stored through the device/TAG mechanism
frame_index	FIELD	INT64	Original per-episode frame index
sample_index	FIELD	INT64	Renamed from source index
action_0 ... action_5	FIELD	FLOAT	Flattened source action[6]
observation_state_0 ... observation_state_5	FIELD	FLOAT	Flattened source observation.state[6]

Vector prefixes preserve the source column names, with . replaced by _. The source timestamp is removed after Time synthesis because it equals Time / 1000 seconds. No frame rows, action/state elements, frame indexes, episode indexes, or task indexes are removed. Camera pixels are external video assets and were never fields in the source Parquet tables.

Storage profile

• Time: TS_2DIFF + LZ4
• FLOAT/DOUBLE: GORILLA + LZ4
• INT32/INT64: TS_2DIFF + LZ4
• BOOLEAN: RLE + LZ4 when present; this source has no BOOLEAN field
• TAG values: TsFile table-model device/tag storage

All episodes are merged into one table and sorted by episode_index, task_index, and Time.

Source videos

Videos are not included. The original repository stores 202 frame-aligned AV1 MP4 files under videos/: 101 files for observation.images.webcam and 101 files for observation.images.wrist. The path template is videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4. The video files total 684,781,139 bytes, are 640x480 at 30 fps, and have no audio. Use episode_index and frame_index to align a row with both views.

Data checks

The generated TsFile was read in full with the Apache TsFile Java reader. Source, staged, and TsFile row counts all equal 31,354. Physical column codecs, TAG roles, vector widths, per-episode Time ordering, duplicate TAG+Time keys, and file-size reduction were also checked locally.

Minimal read example

from tsfile import TsFileReader

reader = TsFileReader("data/jclinton1_wedgit_stack_single_dual_cam.tsfile")
columns = [
    "episode_index",
    "task_index",
    "frame_index",
    "sample_index",
    "action_0",
    "observation_state_0",
]
with reader.query_table("jclinton1_wedgit_stack_single_dual_cam", columns, batch_size=65536) as result:
    batch = result.read_arrow_batch()
    print(batch.to_pandas().head())
reader.close()

When using this dataset, cite the original Hugging Face repository and Joe Clinton (jclinton1).