THULab/berkeley_cable_routing

berkeleycablerouting

This dataset is a TsFile conversion of the Hugging Face dataset `lerobot/berkeley_cable_routing`. The source dataset was created using LeRobot.

Modalities: Time-series. The original repository also contains video streams; videos are not included in this converted repository.

Source Dataset

• —Original dataset: `lerobot/berkeley_cable_routing`
• —License: apache-2.0
• —LeRobot codebase version: v3.0
• —Robot type: unknown
• —Task: route cable
• —Split: train (0:1647)
• —Source scale from meta/info.json: 1,647 episodes, 42,328 frames, 1 task
• —Sampling rate: 10 fps
• —Source data 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/berkeley_cable_routing.tsfile
• —Converted rows: 42,328
• —TsFile table: berkeley_cable_routing
• —Time precision: milliseconds
• —TAG columns: episode_index, task_index

The conversion follows the LeRobot v3 data path in meta/info.json and uses the file-*.parquet frame data layout.

Schema

Time is synthesized as round(timestamp * 1000) in milliseconds. The source timestamp column is dropped because it is redundant with Time / 1000 seconds.

TAG columns:

• —episode_index
• —task_index

FIELD columns:

• —frame_index
• —sample_index (renamed from source index)
• —next_reward (renamed from source next.reward)
• —next_done (renamed from source next.done, stored as 0/1)
• —observation_state_0 to observation_state_7
• —action_0 to action_6

Vector features are flattened by preserving the source feature name and replacing . with _. For example, observation.state becomes observation_state_0 to observation_state_7, and action becomes action_0 to action_6.

Video Policy

The following source video features are not converted into TsFile and are not uploaded here:

• —observation.images.top_image
• —observation.images.wrist225_image
• —observation.images.wrist45_image
• —observation.images.image

Use the original dataset for videos: `lerobot/berkeley_cable_routing/videos`.

Validation

The converted TsFile was validated with the project pipeline and read back using the TsFile Python SDK:

• —staged Parquet rows: 42,328
• —TsFile metadata rows: 42,328
• —TsFile query rows: 42,328

Usage

from tsfile import TsFileReader

path = "data/berkeley_cable_routing.tsfile"
with TsFileReader(path) as reader:
    schemas = reader.get_all_table_schemas()
    print(schemas.keys())