# Qiyuan Robotics rolls out 5,000 robots, discusses fast-slow brain division of labor and Hil-UMI data paradigm

- Source: 新智元 · 微信公众号
- Published: 2026-10-29T12:55:58.000Z
- RoboHot: https://robohot.news/en/items/u340i2qld4xtsq6ep67p06uxj
- Original: https://mp.weixin.qq.com/s?__biz=MzI3MTA0MTk1MA==&mid=2652739039&idx=1&sn=6f16c7d394f58f1c3aac4d62bcee5967&chksm=f0acf5c7d7f1b3ee4bca52daaeafd5d12c097ad618a178cd53f28859a95944d0eb09dc752bef&scene=126&sessionid=1793283320#rd

## Summary

Qiyuan Robotics announced it has rolled out 5,000 units and disclosed its Physical AI technology roadmap: a GPT-type large model serves as the S2 slow brain for high-level planning, while its self-developed VLA acts as the S1 fast brain to execute tasks such as folding clothes and tidying desktops. Its newly proposed Hil-UMI collection paradigm achieved data collection efficiency 5.63 times that of conventional real-robot HG-DAgger in a desktop-tidying experiment. The team believes general large models are a public resource, with the moat lying in the robot body, high-quality data, and the training loop, and says consumer household chores need task completion rates above 95% before they meet commercialization conditions.