Skip to content

unienv_data.integrations.lerobot

Classes:

Name Description
LeRobotAsUniEnvDataset

Wraps a local LeRobot dataset directory as a UniEnv BatchBase.

LeRobotAsUniEnvDataset

LeRobotAsUniEnvDataset(
    dataset_dir: str,
    schema: LeRobotSchema,
    index: LeRobotIndex,
    space: DictSpace,
    _backend: ComputeBackend,
    _device: Optional[Any] = None,
    decode_video: bool = True,
    video_backend: Literal["pyav", "torchcodec", "auto"] = "auto",
    include_features: Optional[List[str]] = None,
    exclude_features: Optional[List[str]] = None,
    video_hwaccel: Optional[Any] = "auto",
)

Bases: BatchBase[BatchT, BArrayType, BDeviceType, BDtypeType, BRNGType]

Wraps a local LeRobot dataset directory as a UniEnv BatchBase.

Parses the LeRobot on-disk format natively (no lerobot package required). Supports v2.0, v2.1, and v3.0 dataset formats.

Usage::

from unienv_interface.backends.numpy import NumpyComputeBackend
dataset = LeRobotAsUniEnvDataset.from_local(
    "/path/to/lerobot/dataset",
    backend=NumpyComputeBackend,
)
frame = dataset[42]
batch = dataset[0:100]

Methods:

Name Description
from_local

Load a LeRobot dataset from a local directory.

from_hub

Download a LeRobot dataset from HuggingFace Hub and load it.

get_at
get_at_with_metadata
get_column

Return a view over a single feature column, joining nested keys with dots.

get_episode

Load all frames for a single episode.

close
get_flattened_at

Fetch samples as flattened backend arrays.

get_flattened_at_with_metadata

Fetch flattened samples together with optional per-sample metadata.

set_flattened_at

Overwrite existing samples using flattened data.

append_flattened

Append one flattened sample to the batch.

extend_flattened

Append a batched block of flattened samples.

set_at

Overwrite existing samples using structured data.

remove_at

Remove one or more samples from the batch.

append

Append one structured sample to the batch.

extend

Append a batched block of structured samples.

extend_from

Copy data from another batch in bounded-size chunks.

get_slice

Create a lazy view over a subset of indices.

Attributes:

Name Type Description
is_mutable
num_episodes int
fps int
schema LeRobotSchema
episode_metadata List[EpisodeMetadata]
single_space
single_metadata_space
backend ComputeBackend[BArrayType, BDeviceType, BDtypeType, BRNGType]
device Optional[BDeviceType]

is_mutable class-attribute instance-attribute

is_mutable = False

num_episodes property

num_episodes: int

fps property

fps: int

schema property

schema: LeRobotSchema

episode_metadata property

episode_metadata: List[EpisodeMetadata]

single_space instance-attribute

single_space = single_space

single_metadata_space instance-attribute

single_metadata_space = single_metadata_space

backend property

backend: ComputeBackend[BArrayType, BDeviceType, BDtypeType, BRNGType]

device property

device: Optional[BDeviceType]

from_local staticmethod

from_local(
    dataset_dir: str,
    backend: ComputeBackend[BArrayType, BDeviceType, BDtypeType, BRNGType],
    device: Optional[BDeviceType] = None,
    include_features: Optional[List[str]] = None,
    exclude_features: Optional[List[str]] = None,
    decode_video: bool = True,
    video_backend: Literal["pyav", "torchcodec", "auto"] = "auto",
    video_hwaccel: Optional[Any] = "auto",
) -> LeRobotAsUniEnvDataset[BArrayType, BDeviceType, BDtypeType, BRNGType]

Load a LeRobot dataset from a local directory.

video_hwaccel is passed to the video readers: "auto" (default) uses a hardware decoder when one is available, None decodes in software, which is cheaper to open and often faster for small frames read at random positions (the usual training case).

from_hub staticmethod

from_hub(
    repo_id: str,
    backend: ComputeBackend[BArrayType, BDeviceType, BDtypeType, BRNGType],
    device: Optional[BDeviceType] = None,
    revision: Optional[str] = None,
    local_dir: Optional[str] = None,
    **kwargs,
) -> LeRobotAsUniEnvDataset[BArrayType, BDeviceType, BDtypeType, BRNGType]

Download a LeRobot dataset from HuggingFace Hub and load it.

Uses huggingface_hub.snapshot_download (not the lerobot package).

get_at

get_at(idx: Union[IndexableType, BArrayType]) -> BatchT

get_at_with_metadata

get_at_with_metadata(
    idx: Union[IndexableType, BArrayType],
) -> Tuple[BatchT, Dict[str, Any]]

get_column

get_column(nested_keys: Sequence[str]) -> BatchBase

Return a view over a single feature column, joining nested keys with dots.

get_episode

get_episode(episode_index: int) -> BatchT

Load all frames for a single episode.

close

close() -> None

get_flattened_at

get_flattened_at(idx: Union[IndexableType, BArrayType]) -> BArrayType

Fetch samples as flattened backend arrays.

get_flattened_at_with_metadata

get_flattened_at_with_metadata(
    idx: Union[IndexableType, BArrayType],
) -> Tuple[BArrayType, Optional[Dict[str, Any]]]

Fetch flattened samples together with optional per-sample metadata.

set_flattened_at

set_flattened_at(idx: Union[IndexableType, BArrayType], value: BArrayType) -> None

Overwrite existing samples using flattened data.

append_flattened

append_flattened(value: BArrayType) -> None

Append one flattened sample to the batch.

extend_flattened

extend_flattened(value: BArrayType) -> None

Append a batched block of flattened samples.

set_at

set_at(idx: Union[IndexableType, BArrayType], value: BatchT) -> None

Overwrite existing samples using structured data.

remove_at

remove_at(idx: Union[IndexableType, BArrayType]) -> None

Remove one or more samples from the batch.

append

append(value: BatchT) -> None

Append one structured sample to the batch.

extend

extend(value: BatchT) -> None

Append a batched block of structured samples.

extend_from

extend_from(
    other: BatchBase[BatchT, BArrayType, BDeviceType, BDtypeType, BRNGType],
    chunk_size: int = 8,
    tqdm: bool = False,
) -> None

Copy data from another batch in bounded-size chunks.

get_slice

get_slice(
    idx: Union[IndexableType, BArrayType],
) -> BatchBase[BatchT, BArrayType, BDeviceType, BDtypeType, BRNGType]

Create a lazy view over a subset of indices.