Founded in 2024, the company builds vision language action models for robots and open-sourced its Spirit v1.5 system in January.