THULab/johnsutor_MuJoCoPickAndPlace-v1

MuJoCoPickAndPlace v1 TsFile

Apache TsFile representation of johnsutor/MuJoCoPickAndPlace-v1, a LeRobot v3.0 MuJoCo robotics dataset.

Source

• Author and publisher: John Sutor (johnsutor)
• License: Apache-2.0
• Task: Pick up the red cube and place it on the blue circle.
• Robot type: SO-101
• Split: train (0:10)
• Sampling rate: 30 fps
• Episodes: 10; frame rows: 4,725; tasks: 1; source Parquet shards: 1
• Source frame layout: data/chunk-{chunk_index:03d}/file-{file_index:03d}.parquet

The source card does not provide a paper or completed BibTeX citation.

TsFile layout

The numeric rows are stored in data/johnsutor_mujoco_pick_and_place_v1.tsfile as table johnsutor_mujoco_pick_and_place_v1. Source metadata JSON is retained under meta/; the task and episode metadata are represented as JSONL, and no source Parquet file is placed there.

Time = round(timestamp * 1000) is an INT64 millisecond timeline that restarts at zero for every episode. The source timestamp is omitted because it equals Time / 1000 seconds. index is renamed to sample_index, and frame_index is preserved.

Role	Columns
TIME	Time (INT64, milliseconds)
TAG/device	episode_index, task_index
FIELD	frame_index, sample_index (INT64)
FIELD	action[6] -> action_0 ... action_5 (FLOAT)
FIELD	observation.state[6] -> observation_state_0 ... observation_state_5 (FLOAT)
FIELD	observation.environment_state[36] -> observation_environment_state_0 ... observation_environment_state_35 (FLOAT)
FIELD	reward, success, done, and six reward_components_* values (FLOAT)

No numeric row, action dimension, state dimension, environment-state dimension, reward value, or frame index is dropped.

Videos

Videos remain in the original Hugging Face repository and are not included here:

• videos/observation.images.overhead/chunk-000/file-000.mp4 (23,757,203 bytes)
• videos/observation.images.wrist/chunk-000/file-000.mp4 (17,189,090 bytes)

The source layout is videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4. Numeric rows align with both camera streams through episode_index, frame_index, and the offsets recorded in meta/episodes.jsonl.

Read example

from tsfile import TsFileReader

reader = TsFileReader("data/johnsutor_mujoco_pick_and_place_v1.tsfile")
columns = [
    "episode_index",
    "task_index",
    "frame_index",
    "sample_index",
    "action_0",
    "observation_state_0",
    "observation_environment_state_0",
    "reward",
]
with reader.query_table("johnsutor_mujoco_pick_and_place_v1", columns, batch_size=4096) as result:
    print(result.read_arrow_batch().to_pandas().head())
reader.close()