THULab/alicia-dataset-test

Alicia Dataset Test TsFile

This dataset is a TsFile conversion of scaling-robotics/alicia-dataset-test. It stores the numeric LeRobot frame data as an Apache TsFile table while keeping source metadata under meta/. Camera videos are not included in this converted repository.

Modalities: Time-series.

Source Dataset

• Source: scaling-robotics/alicia-dataset-test
• Source owner: scaling-robotics
• Source format: LeRobot v3.0
• Robot type: alicia_d_follower
• Split: train
• Episodes: 501
• Frames / TsFile rows: 185,137
• Tasks: 1
• Sampling rate: 30 fps

Converted Files

• data/alicia_dataset_test.tsfile: converted numeric time-series data.
• meta/: source metadata mirrored from the original dataset, with meta/info.json updated to describe the converted TsFile schema and record the original video paths.

TsFile Schema

Table name: alicia_dataset_test

Time:

• Time: INT64 milliseconds, computed as round(timestamp * 1000).

TAG columns:

• episode_index: source episode index.
• task_index: source task index.

FIELD columns:

• frame_index: source frame index.
• sample_index: renamed from source index.
• action_0 through action_6: flattened source action vector, FLOAT.
• observation_state_0 through observation_state_6: flattened source observation.state vector, FLOAT.

Conversion Notes

• Converted with the generic lerobot converter in the huggingface_to_tsfile workflow.
• All episodes are stored in one TsFile table. Use episode_index and task_index TAG filters to select an episode or task.
• Source vector columns are flattened by preserving the source column name, replacing . with _, and appending the element index.
• The source timestamp column is not retained as a separate field because it is redundant with Time / 1000 seconds.
• Source index is renamed to sample_index.
• Video features are omitted from the converted repository: observation.images.front_camera and observation.images.wrist_camera. Videos remain in the original dataset under videos/.

Validation

The staged Parquet and generated TsFile were checked locally:

• Staged Parquet rows: 185,137
• TsFile metadata rows: 185,137
• TsFile query readback rows: 185,137
• Generated TsFile size: 3,992,560 bytes

Minimal Read Example

from tsfile import TsFileReader

path = "data/alicia_dataset_test.tsfile"
reader = TsFileReader(path)
schemas = reader.get_all_table_schemas()
print(schemas.keys())