Create a LeRobot Dataset

SUMMARY

Use LeRobotDataset.create() to define a new LeRobot v3 dataset and return a writable dataset ready to receive frames and episodes.

Learn about LeRobot datasets

Review the dataset structure, feature schema, storage layout, and workflow.

View overview →

Method

Use LeRobotDataset.create() to initialize a writable dataset:

from telekinesis.dataengine import datasets

dataset = datasets.LeRobotDataset.create(
    repo_id="user/my_dataset",
    local_path="results/user/my_dataset",
    fps=30,
    features=features,
)

This is the lower-level dataset creation workflow. After creation, you can write frames and episodes directly to the dataset or use LeRobotDatasetLogger when you want a managed recording lifecycle.

Parameter Configuration

Parameter	Type	Default	Description
repo_id	str	required	Repository identifier, usually organization/dataset.
fps	int	required	Capture frame rate written to metadata.
features	dict	required	Feature names mapped to dtype, shape, and dimension names.
local_path	str | Path | None	None	Dataset directory; otherwise uses the default LeRobot home.
robot_type	str | None	None	Robot name stored in metadata.
use_videos	bool	True	Encode visual streams as video instead of individual images.

Writer and encoder tuning parameters are listed in Configuration.

Feature Schema

features maps each recorded feature key to its data type, shape, and optional dimension names. Every frame added to the dataset must match this schema.

features = {
    "observation.images.camera1": {
        "dtype": "video",
        "shape": [64, 64, 3],
        "names": ["height", "width", "channels"],
    },
    "observation.state": {
        "dtype": "float32",
        "shape": [6],
        "names": None,
    },
    "action": {
        "dtype": "float32",
        "shape": [6],
        "names": None,
    },
}

See the Feature Schema reference for complete understanding and naming conventions.

Example

"""Example script demonstrating how to create a LeRobot dataset using the Telekinesis Data Engine."""

from pathlib import Path
import shutil

from loguru import logger

from telekinesis.dataengine import datasets


def create_lerobot_dataset_example():
    """Create a new writable LeRobot dataset."""

    # 1. Define the dataset identity, local storage path, and features.
    repo_id = "user/my_create_example"
    local_path = (
        Path(__file__).resolve().parent.parent.parent.parent
        / "results"
        / repo_id
    )
    # Remove any previous example dataset so this script can be rerun.
    if local_path.exists():
        shutil.rmtree(local_path)

    features = {
        "observation.images.camera1": {
            "dtype": "video",
            "shape": [480, 640, 3],
            "names": ["height", "width", "channels"],
        },
        "observation.depths.camera1": {
            "dtype": "depth",
            "shape": [480, 640],
            "names": ["height", "width"],
        },
        "observation.state": {
            "dtype": "float32",
            "shape": [6],
            "names": [
                "shoulder_pan_joint.pos",
                "shoulder_lift_joint.pos",
                "elbow_joint.pos",
                "wrist_1_joint.pos",
                "wrist_2_joint.pos",
                "wrist_3_joint.pos",
            ],
        },
        "action": {
            "dtype": "float32",
            "shape": [6],
            "names": [
                "shoulder_pan_joint.pos",
                "shoulder_lift_joint.pos",
                "elbow_joint.pos",
                "wrist_1_joint.pos",
                "wrist_2_joint.pos",
                "wrist_3_joint.pos",
            ],
        },
        "language_persistent": {
            "dtype": "text",
            "shape": [],
            "names": None,
        },
    }

    # 2. Create the LeRobot dataset.
    dataset = datasets.LeRobotDataset.create(
        repo_id=repo_id,
        local_path=local_path,
        fps=30,
        features=features,
        robot_type="dummy_ur10e",
        use_videos=True,
    )

    logger.info("LeRobot dataset created successfully.")
    logger.info(dataset)
    logger.info(f"Local path: {local_path}")

    # 3. Finalize the dataset when no more data will be written.
    dataset.finalize()

    return dataset


if __name__ == "__main__":
    create_lerobot_dataset_example()

Next Steps

Write to a LeRobot Dataset

Add frames, save episodes, and finalize the writable dataset.

Write data →

Configure Dataset Writing

Configure buffering, video encoding, streaming, and shard sizes.

View configuration →

Record with the Dataset Logger

Use the managed episode lifecycle for live robot data collection.

Start recording →

Inspect Dataset Metadata

Review the dataset schema, frame rate, episodes, statistics, and storage metadata.

Inspect metadata →