Documentation preview

Smiling Buddha is an open research initiative building the real-world data layer for physical AI. This preview outlines the core concepts and data model. The full SDK, CLI, and documentation site will launch alongside the first public preview dataset.

Data model

All Smiling Buddha datasets will follow the LeRobot v2.1 native format. This gives you three things:

Immediate compatibility with the lerobot Python library
Consistent episode / observation / action structure across all datasets
Version-safe schema evolution — old datasets keep working with new library versions

V0 → V3 validation taxonomy

Every artifact carries a validation score from V0 (synthetic base, no real-world anchor) to V3 (novel discovery from live deployment). These definitions are the canonical version and match the working paper and blog posts:

V0 — Synthetic base. Simulation, video generation, or world-model rollouts with no real-world anchor.

V1 — Single real match. One production observation anchors a set of synthetic variations.

V2 — Multiple real matches. Two or more real cases from different sites, embodiments, or environmental conditions.

V3 — Novel discovery. A real edge case that no prior synthesis produced. Emerges only from live deployment.

LeRobot v2.1 native

Smiling Buddha datasets are LeRobot v2.1 native — the schema evolution, episode metadata layout, and multi-camera observation format all follow the LeRobot conventions directly. Once the preview corpus opens in Q4 2026, migration between versions will be automatic.

Get notified when v0.1 lands

The SDK, CLI, and full documentation will publish together with the first preview corpus. To be notified when the alpha lands: email us with your name and institution.