Struct re_data_loader::lerobot::LeRobotDataset

pub struct LeRobotDataset {
    pub path: PathBuf,
    pub metadata: LeRobotDatasetMetadata,
}

A LeRobot dataset consists of structured metadata and recorded episode data stored in Parquet files.

LeRobot Dataset Format

The dataset follows a standardized directory layout, typically organized as follows:

.
├── README.md
├── data
│  └── chunk-000
│      ├── episode_000000.parquet
│      ├── episode_000001.parquet
│      ├── …
├── meta
│  ├── episodes.jsonl
│  ├── info.json
│  ├── stats.json
│  └── tasks.jsonl
└── videos
    └── chunk-000
        └── observation.image
            ├── episode_000000.mp4
            ├── episode_000001.mp4
            ├── …

File layout

• data/: Stores episode data in Parquet format, organized in chunks.
• meta/: Contains metadata files:
  • info.json: General dataset metadata (robot type, number of episodes, etc.).
  • episodes.jsonl: Episode-specific metadata (tasks, number of frames, etc.).
  • tasks.jsonl: Task definitions for episodes.
  • stats.json: Summary statistics of dataset features.
• videos/: Optional directory storing video observations for episodes, organized similarly to data/.

Each episode is identified by a unique index and mapped to its corresponding chunk, based on the number of episodes per chunk (which can be found in meta/info.json).

Fields

path: PathBuf

metadata: LeRobotDatasetMetadata

Implementations

impl LeRobotDataset

pub fn load_from_directory(path: impl AsRef<Path>) -> Result<Self, LeRobotError>

Loads a LeRobotDataset from a directory.

This method initializes a dataset by reading its metadata from the meta/ directory.

Important

Currently, this only supports v2 LeRobot datasets.

pub fn read_episode_data( &self, episode: EpisodeIndex, ) -> Result<RecordBatch, LeRobotError>

Read the Parquet data file for the provided episode.

pub fn read_episode_video_contents( &self, observation_key: &str, episode: EpisodeIndex, ) -> Result<Cow<'_, [u8]>, LeRobotError>

Read video feature for the provided episode.

pub fn task_by_index(&self, task: TaskIndex) -> Option<&LeRobotDatasetTask>

Retrieve the task using the provided task index.

Trait Implementations

impl Debug for LeRobotDataset

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.