Dataset Operations

SUMMARY

Dataset Operations provide consistent interfaces for creating, loading, writing, inspecting, configuring, and visualizing supported robotics dataset formats.

Explore the Data Engine

See how dataset operations connect data generation, collection, training, deployment, and continuous improvement.

View overview →

When to Use Dataset Operations?

Use Dataset Operations when you need to:

• Load an existing local or remote dataset.
• Create a dataset
• Inspect and analyse datasets.
• Modify frames and images, save episodes and annotations.
• Visualize datasets.

How to Use Dataset Operations?

Dataset workflows vary by format, but generally follow the same lifecycle:

1. Select a dataset format based on the data and intended use.
2. Create a dataset or load data from an existing source.
3. Add, read, organize, or transform its records and assets.
4. Validate data quality, structure, and metadata.
5. Visualize, analyze, export, or use the dataset in a training workflow.

DATA LOGGER OR DATASET OPERATIONS?

Use Data Loggers to capture data during robot execution. Use Dataset Operations to manage and consume the resulting datasets.

Guides

Currently support available for LeRobot dataset format.

INFO

Support for additional formats, including COCO datasets, is coming soon.

LeRobot Dataset

Create, write, load, inspect, configure, and visualize LeRobot v3 datasets for robot learning.

Explore LeRobot →