The brief

Analyst view

• Generalist AI builds robot foundation models trained on diverse manipulation tasks.
• The company raised $400 million in June 2026 with backing from NVIDIA and Bezos Expeditions.
• Its models learn new robot tasks from human demonstrations and can be deployed across different robot bodies.

Technical approach

As reported

Vision-language-action models

Learning from human demonstrations

Models that watch human performing a task and learn the relationship between visual input, language commands and robot actions.

Generalization across tasks

One model for many uses

The goal is a foundation model that handles new manipulation tasks without retraining and works across different robot bodies.