myarm-3-put-cube-in-basket-highres (LeRobot) — TsFile

This dataset converts the numeric time-series of the HuggingFace LeRobot dataset apockill/myarm-3-put-cube-in-basket-highres to the Apache TsFile format, one TsFile per episode, and keeps the original camera videos alongside.

Original dataset

• Source dataset: apockill/myarm-3-put-cube-in-basket-highres
• Format: LeRobot v2.0 (robot_type = myarm)
• Content: a myarm robot-arm manipulation dataset — 15,414 frames / 54 episodes / 20 fps, task: "Look for and pick up a cube and then look for and place it into the black bin.", with one wrist RGB camera stream.

What is in this repository

data/chunk-000/
└── episode_000000.tsfile … episode_000053.tsfile   # 54 TsFiles (one per episode)
videos/chunk-000/observation.images.wrist/
└── episode_000000.mp4 … episode_000053.mp4         # copied verbatim
meta/
├── info.json        # original LeRobot info + tsfile_conversion notes
├── tasks.jsonl      # copied verbatim
├── episodes.jsonl   # copied verbatim
└── stats.json       # copied verbatim

One TsFile per episode (1:1 with the source)

Unlike a single merged table, each source episode_NNNNNN.parquet is converted to its own episode_NNNNNN.tsfile at the same path. This preserves the per-episode layout and aligns 1:1 with the 54 per-episode MP4 videos.

TsFile storage mapping (table model, per file)

Role	Column(s)	Type	Notes
TAG	episode_id	STRING	episode_NNNNNN, aligned with the filename; constant within the file
TAG	task_id	STRING	task_K (from task_index); constant within the file
Time	frame_index * 50 ms	INT64 (ms)	20 fps → exact 50 ms interval
FIELD	state_0 … state_6	FLOAT	observation.state[7] expanded
FIELD	action_0 … action_6	FLOAT	action[7] expanded
FIELD	episode_index, frame_index, sample_index, task_index	INT64	indices (index → sample_index)
FIELD	episode_timestamp_s	FLOAT	(timestamp)

Each file holds a single episode, so the two TAG columns are constant within a file — they are kept so every file shares the same table schema and can be queried the same way (e.g. filter by episode_id / task_id).

Conversion notes

• One TsFile per episode — 54 source parquet → 54 TsFiles, same paths/names.
• Only the numeric time-series is converted; the observation.images.wrist pixels live in the MP4 files under videos/, kept verbatim here.
• TAG = episode_id + task_id (constant within each single-episode file, retained for a uniform schema across the 54 files).
• Time = frame_index × 50 ms (20 fps → exact 50 ms, no rounding). frame_index is strictly increasing within each episode.
• Array columns expanded: observation.state[7] → state_0..state_6, action[7] → action_0..action_6 (FLOAT).
• No columns dropped, no rows dropped: all 15,414 frames preserved across the 54 files.

Usage

from tsfile import TsFileReader

reader = TsFileReader("data/chunk-000/episode_000000.tsfile")
schemas = reader.get_all_table_schemas()
tname = next(iter(schemas))

cols = ["state_0", "action_0", "frame_index", "task_index"]
with reader.query_table(tname, cols, batch_size=65536) as rs:
    while (batch := rs.read_arrow_batch()) is not None:
        df = batch.to_pandas()
        # ... process ...
reader.close()

Citation

@misc{myarm_put_cube_highres,
  title  = {myarm-3-put-cube-in-basket-highres (LeRobot)},
  author = {apockill},
  url    = {https://huggingface.co/datasets/apockill/myarm-3-put-cube-in-basket-highres},
  publisher = {Hugging Face}
}

The source HuggingFace dataset does not declare an explicit license.