Inspect LeRobot Metadata

SUMMARY

Use LeRobotDatasetMetadata to inspect a LeRobot dataset's schema, frame rate, robot type, episodes, statistics, and storage layout without manually parsing its metadata files.

Learn about LeRobot datasets

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

View overview →

Method

Create LeRobotDatasetMetadata with the existing dataset's repository ID and local path:

from telekinesis.dataengine import datasets

metadata = datasets.LeRobotDatasetMetadata(
    repo_id="lerobot/aloha_sim_insertion_scripted",
    local_path="results/lerobot/aloha_sim_insertion_scripted",
)

Parameter Configuration

Parameter	Type	Default	Description
repo_id	str	required	Repository identifier of the dataset to inspect.
local_path	str | Path | None	None	Local dataset directory; otherwise uses the default LeRobot home.
revision	str | None	None	Hub branch, tag, or commit to inspect.
force_cache_sync	bool	False	Refresh cached metadata from the Hub.

Attribute Reference

Property	Type	Description
fps	int	Collection frame rate.
version	str	LeRobot codebase version recorded by the dataset.
robot_type	str | None	Robot type stored in the dataset metadata.
total_episodes	int	Total number of recorded episodes.
total_frames	int	Total number of recorded frames.
total_tasks	int	Total number of task definitions.
features	dict[str, dict]	Dataset feature schema.
names	dict[str, list | dict]	Dimension or element names for each feature.
shapes	dict[str, tuple]	Shape of each dataset feature.
image_keys	list[str]	Visual feature keys stored as images or videos.
video_keys	list[str]	Visual feature keys stored as videos.
depth_keys	list[str]	Feature keys stored as depth maps.
camera_keys	list[str]	Camera feature keys stored as images or videos.
has_language_columns	bool	Whether the dataset contains a language feature.
episodes	dataset table	Episode metadata records.
tasks	dataset table	Task metadata records.
stats	dict | None	Stored dataset statistics.
tools	list[dict]	Tool schemas declared by the dataset.

Indexing Reference

Method	Return Type	Description
get_data_file_path(ep_index)	Path	Returns the relative Parquet path for a zero-based episode index.
get_video_file_path(ep_index, vid_key)	Path	Returns the relative video path for an episode index and video feature key.
get_task_index(task)	int | None	Returns the index assigned to a task description, or None if the task is not registered.

Example

"""Example script demonstrating how to inspect LeRobot dataset metadata."""

from pathlib import Path

from loguru import logger

from telekinesis.dataengine import datasets

def inspect_lerobot_metadata_example():
    """Inspect metadata of an existing LeRobot dataset."""

    # 1. Define the dataset identity and local path.
    repo_id = "lerobot/aloha_sim_insertion_scripted"
    local_path = (
        Path(__file__).resolve().parent.parent.parent.parent
        / "results"
        / repo_id
    )

    # 2. Load the dataset metadata.
    metadata = datasets.LeRobotDatasetMetadata(
        repo_id=repo_id,
        local_path=local_path,
    )

    logger.info("Metadata loaded successfully.")
    logger.info(metadata)

    # 3. Inspect general dataset information.
    logger.info(f"Local path: {local_path}")
    logger.info(f"URL root: {metadata.url_root}")
    logger.info(f"Codebase version: {metadata.version}")
    logger.info(f"Robot type: {metadata.robot_type}")
    logger.info(f"FPS: {metadata.fps}")

    logger.info(f"Data path: {metadata.data_path}")
    logger.info(f"Video path: {metadata.video_path}")

    # 4. Inspect the dataset schema and features.
    logger.info(f"Features: {metadata.features}")
    logger.info(f"Names: {metadata.names}")
    logger.info(f"Shapes: {metadata.shapes}")

    logger.info(f"Image keys: {metadata.image_keys}")
    logger.info(f"Video keys: {metadata.video_keys}")
    logger.info(f"Depth keys: {metadata.depth_keys}")
    logger.info(f"Camera keys: {metadata.camera_keys}")
    logger.info(f"Has language columns: {metadata.has_language_columns}")

    # 5. Inspect dataset statistics and episode information.
    logger.info(f"Total episodes: {metadata.total_episodes}")
    logger.info(f"Total frames: {metadata.total_frames}")
    logger.info(f"Total tasks: {metadata.total_tasks}")

    logger.info(f"Tasks: {metadata.tasks}")
    logger.info(f"Episodes: {metadata.episodes}")
    logger.info(f"Statistics: {metadata.stats}")
    logger.info(f"Tools: {metadata.tools}")

    # 6. Inspect dataset storage and chunking configuration.
    logger.info(f"Max chunk size: {metadata.max_chunk_size}")
    logger.info(
        f"Max data file size (MB): {metadata.max_data_files_size_in_mb}",
    )
    logger.info(f"Max video file size (MB): {metadata.max_video_files_size_in_mb}")
    logger.info(f"Chunk settings: {metadata.get_chunk_settings()}")

    # 7. Resolve metadata for a specific episode.
    episode_index = 0
    # Use the key as per the lerobot features downloaded
    video_key = "observation.images.top"

    logger.info(
        f"Episode {episode_index} data path: {metadata.get_data_file_path(episode_index)}"
    )
    logger.info(
        f"Episode {episode_index} video path for '{video_key}': {metadata.get_video_file_path(episode_index, video_key)}"
    )


if __name__ == "__main__":
    inspect_lerobot_metadata_example()

Next Steps

Load the Dataset

Open selected episodes and access decoded frame data.

Load dataset →

Visualize the Dataset

Inspect synchronized observations and actions in Rerun.

Visualize dataset →

LeRobotDataset Overview

Explore all supported dataset creation, writing, loading, and inspection workflows.

View overview →