THULab/recreate-bug-pre-fix-v1-trim

Recreate Bug Pre Fix V1 Trim TsFile

This dataset is an Apache TsFile conversion of VibeCuisine/recreate-bug-pre-fix-v1-trim, a LeRobot v3 robot-manipulation dataset.

Modalities: Time-series. The converted repository contains numeric robot observations, actions, auxiliary controller signals, frame timing, episode/task tags, and mirrored source metadata. Camera videos remain in the original Hugging Face dataset.

Source Dataset

• Source dataset: VibeCuisine/recreate-bug-pre-fix-v1-trim
• Original dataset author/uploader shown in the Hugging Face repository history: avu121
• Source organization: VibeCuisine
• License: Apache-2.0
• Robot type: vibeboard_follower_tilt
• LeRobot codebase version: v3.0
• Task: Grab the cucumber close to one of the cucumber's ends.
• Split: train
• Scale: 181 episodes, 15,317 frames, 1 task, 20 fps
• Source frame shards: 1
• Source frame 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

Converted Files

• TsFile: data/recreate_bug_pre_fix_v1_trim_train.tsfile
• Table: recreate_bug_pre_fix_v1_trim_train
• Rows: 15,317
• Episodes: 181
• Time precision: milliseconds
• Metadata: meta/ is mirrored from the source, with meta/info.json updated to describe the TsFile artifact.

Schema

Time is synthesized as round(timestamp * 1000) milliseconds and restarts for each episode.

TAG columns:

• episode_index
• task_index

Scalar FIELD columns:

• frame_index
• sample_index, renamed from source index
• aux_elevator_mm, renamed from aux.elevator_mm
• aux_spinner_active, renamed from aux.spinner_active
• aux_limit_home, renamed from aux.limit_home
• aux_limit_gripper, renamed from aux.limit_gripper

Flattened FLOAT FIELD groups:

• action[7] -> action_0 ... action_6
• observation.state[7] -> observation_state_0 ... observation_state_6

Conversion Notes

• The shared config-driven lerobot converter was used; the dataset-specific script is a thin local orchestration entry point.
• The train split is merged into one table-model TsFile. Use episode_index and task_index TAG filters to select an episode or task.
• Vector columns are flattened to scalar TsFile fields. Full source prefixes are preserved, with . replaced by _.
• The source timestamp column is dropped after Time synthesis because it is redundant with Time / 1000 seconds.
• Source index is renamed to sample_index; the four aux.* scalar columns are retained with dots replaced by underscores.
• No numeric rows or numeric dimensions are intentionally dropped.

TsFile Encoding

• FLOAT/DOUBLE: GORILLA encoding with LZ4 compression.
• INT32/INT64: TS_2DIFF encoding with LZ4 compression.
• Time: TS_2DIFF encoding with LZ4 compression.
• BOOLEAN: configured as RLE with LZ4 when present; these datasets contain no BOOLEAN fields.
• episode_index and task_index are stored as TsFile table-model TAG/device segments.

Videos

Videos are not duplicated in this converted repository. The original videos/ tree is 678 MB and contains three frame-aligned camera streams:

• observation.images.base
• observation.images.top
• observation.images.wrist

The source path template is videos/{video_key}/chunk-{chunk_index:03d}/file-{file_index:03d}.mp4. The numeric rows remain aligned to the original video frames by episode_index, frame_index, and the source episode metadata under meta/episodes/.

Validation

The generated TsFile is checked for successful tool completion, non-zero size, table schema, and row-count equality against the staged Parquet. Expected row count: 15,317.

Usage

from tsfile import TsFileReader

path = "data/recreate_bug_pre_fix_v1_trim_train.tsfile"
reader = TsFileReader(path)
table_name = "recreate_bug_pre_fix_v1_trim_train"
columns = [
    "episode_index",
    "task_index",
    "frame_index",
    "sample_index",
    "action_0",
    "observation_state_0",
]

with reader.query_table(table_name, columns, batch_size=65536) as result:
    batch = result.read_arrow_batch()
    print(batch.to_pandas().head())

Source and License

The source dataset card states that the dataset was created with LeRobot and is licensed under Apache-2.0. The original dataset author/uploader shown in the Hugging Face repository history is avu121, under the VibeCuisine organization. The source card does not provide a paper or completed citation.