Beijing – Spirit AI, a 300‑person Chinese embodied‑intelligence startup, told Reuters its humanoid robot brains are on track for a breakthrough comparable to OpenAI’s GPT‑3 model by mid‑2027. Co‑founder and chief scientist Gao Yang said the “brain is indeed the weakest link in the complete robotics stack,” and that progress is moving quickly.
Industrial rollout before home use
Gao expects the next one to two years to mark the initial window for industrial applications. Within two years, Spirit AI’s robots should appear in commercial service settings performing simpler tasks, while entering homes will remain far harder and likely take at least eight more years.
Spirit AI already has dozens of its Moz1 wheeled humanoid robots operating on production lines at battery maker CATL and retailer JD.com, which also backs the startup. The company reports a 90% success rate for simple tasks in structured living‑room environments.
Data‑driven approach
Unlike many competitors that rely on virtual simulations to cut training costs, Spirit AI trains its robot brains on real‑world data collected by about 1,000 contractors across China. In a Beijing data‑training centre, young workers wear sensors while repeating motions such as opening fridges, unlocking safes and cutting vegetables. The firm found that using “dirty data” – a broader, messier set of motions – helps its models improve faster.
Gao noted that simulators handle rigid bodies well but still struggle with flexible objects like deformable electric cables, underscoring the need for real‑world data.
Funding and outlook
Since its 2024 founding, Spirit AI has raised over $670 million and is valued at roughly 20 billion yuan ($2.9 billion). While the company declined to discuss any plans for an initial public offering, Gao emphasized that early robots could only perform isolated tasks such as pouring water or folding clothing. Today, Spirit AI’s robots can operate across larger spaces and execute continuous, complex workflows.
Despite the promising advances, Gao cautioned that fine‑motor actions like unscrewing a bottle cap and handling unseen tasks remain challenging. The company’s focus on real‑world data collection and incremental industrial deployments reflects a pragmatic path toward the eventual goal of household robots.
Original reporting: Appleton, WI News Feed (HLL/CB) — read the source article.
