LeRobot
LeRobot dataset collection
Demonstration data packaged for Hugging Face LeRobot. Collect once. Train without a custom converter.
What is a LeRobot dataset?
A LeRobot dataset is episodic robot learning data built for the Hugging Face LeRobot library. Each episode holds camera frames, actions, and the metadata you need for imitation learning. Teams use it so loading and training stay consistent across tasks. We collect the demos and package them so they drop into LeRobot workflows without a custom converter.
Longer explainer: What is a LeRobot dataset?
Open datasets vs custom LeRobot data
| Need | Open corpora | Custom (ROBOTRAIN) |
|---|---|---|
| Your objects and fixtures | Rarely | Yes |
| Exclusive rights | No | Available |
| Pinned LeRobot packaging | Varies | Yes |
| RGB-D to your sensors | Sometimes | Yes |
| NDA sites | No | Yes |
Deliverables
- Episodic structure for LeRobot loaders
- RGB observations; RGB-D when specified
- Action streams and task metadata
- Optional language instructions and success flags
- Loader notes for your train config
Workflow
- 01. Scope
Tasks, sensors, volume, LeRobot version, exclusivity.
- 02. Capture
Egocentric, multi-camera, and/or teleop per brief.
- 03. Validate
Sync checks, quality gates, schema conformance.
- 04. Package
Episodes ready for your LeRobot scripts.
FAQ
Format and customization
- What format does LeRobot use?
LeRobot expects episodic trajectories with observations, actions, and metadata its loaders can read. The on-disk layout follows current Hugging Face LeRobot conventions (Parquet episode metadata + MP4 video files + JSON specs). We package natively to your LeRobot dataset version.
- Can datasets be customized?
Yes. You can set tasks, objects, environments, cameras, depth, action schemas, language labels, splits, formats, and licensing. We write a dataset specification before capture starts. Exclusive rights are available. Book a discovery call or ask for a custom quote to lock scope.
- Is LeRobot the same as a robot learning dataset?
LeRobot is the modern open standard for robot learning datasets. A LeRobot dataset is episodic demonstration data structured with Parquet metadata and MP4 videos so LeRobot loaders and training utilities can consume it immediately without custom conversion.
Kit or dataset?
Tell us which. We will reply with next steps.