Spirit AI’s robot-brain roadmap rests on large-scale human data collection
Spirit AI trains its humanoid robot brains on real-world motion data gathered by people rather than in simulation. As of September 2026 the company said it employed around 1,000 contractors across China using wearable data-collection equipment in households and on production lines, and its Beijing offices include a robot data training centre where staff repeatedly carried out household tasks to generate training data. The company says more varied data improved its models faster than clean repeated motions. Co-founder and chief scientist Gao Yang said robot brains face a data bottleneck and that an industry breakthrough comparable to GPT-3.0 could arrive around mid-2027, with home deployment at least eight years away. The company reported a 90% success rate for simple tasks in structured living-room environments.