THULab/Jeevesh2009_so101_gray_block_midvar_test

SO101 Gray Block Midvar Test TsFile

Apache TsFile edition of Jeevesh2009/so101_gray_block_midvar_test, a LeRobot v2.1 SO101 dataset for picking up a gray block and placing it in a box. The numeric trajectories are stored in one table-model TsFile.

Source and attribution

• Original author and repository owner: Madala Venkata Renu Jeevesh (Jeevesh2009).
• License: Apache-2.0.
• The source card does not provide a homepage, paper, or completed BibTeX citation.
• Split: train; 100 episodes; 39,235 frame rows; one task; 30 FPS; 100 source Parquet shards.
• The source meta/info.json declares 39,280 frames, while the 100 Parquet shards and meta/episodes.jsonl both contain 39,235; this TsFile follows the actual source rows.
• Task 0: Pick the gray block and keep it in the box.

Data layout

The table is jeevesh2009_so101_gray_block_midvar_test and contains 39,235 rows across 100 TAG devices. The source Parquet shards total 2,211,671 bytes; the TsFile is 760,250 bytes (34.4% of the source Parquet size).

Column	TsFile role	Type	Meaning
Time	TIME	INT64 milliseconds	round(timestamp * 1000), restarting at zero per episode
episode_index	TAG	STRING from source INT64	Source episode identity
task_index	TAG	STRING from source INT64	Source task identity
frame_index	FIELD	INT64	Frame position within the episode
sample_index	FIELD	INT64	Source index, renamed for clarity
action_0 ... action_5	FIELD	FLOAT	Flattened action[6]
observation_state_0 ... observation_state_5	FIELD	FLOAT	Flattened observation.state[6]

timestamp is not retained as a separate FIELD because it is represented by Time / 1000 seconds. The vector column names preserve their source prefixes, with dots changed to underscores. No trajectory row, episode, task, action dimension, or state dimension is removed.

Videos and alignment

The 200 source AV1 videos are not included here. They remain under videos/chunk-000 in the original repository:

• observation.images.side: 100 files
• observation.images.top: 100 files

The source template is videos/chunk-{episode_chunk:03d}/{video_key}/episode_{episode_index:06d}.mp4. Use episode_index and frame_index to align each numeric row with both 30 FPS video streams.

Read example

from tsfile import TsFileReader

reader = TsFileReader("data/jeevesh2009_so101_gray_block_midvar_test.tsfile")
with reader.query_table(
    "jeevesh2009_so101_gray_block_midvar_test",
    ["episode_index", "task_index", "frame_index", "sample_index", "action_0", "observation_state_0"],
    batch_size=1024,
) as result:
    batch = result.read_arrow_batch()
    print(batch.to_pandas().head())
reader.close()