Embra AI

Embra AI is building the data engine for Physical AI — infrastructure that lets robotics and embodied-intelligence teams discover, evaluate, curate, and source multimodal training data across video, depth, LiDAR, IMU, proprioception, and robot states and actions. Contributors submit datasets scored against seven quality criteria and earn USDC rewards verified on Solana, while AI teams browse a curated catalogue filtered by modality, embodiment, and task.

Robotics models are hungry for real-world multimodal data, but most of it sits locked inside individual labs rather than in shared, quality-graded catalogues. Embra is betting that a marketplace with paid contributors and explicit evaluation criteria can unlock supply the way large text corpora once did for language models.

The company was founded in 2026 in San Francisco by Anthony Smith, positioning itself as the data layer beneath the current wave of embodied-intelligence work — and, notably, as a distinct entity from Embra (embra.ai), the AI assistant for professionals.

Embra says its $1M pre-seed is backed by angels with Boston Dynamics, Agility Robotics, and NVIDIA backgrounds — people who understand the data bottlenecks in robotics from the inside. Beyond founder Anthony Smith, further team pedigree isn't yet public.