Article

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

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.

Part of the event: Qiyuan Robotics Rolls Out Its 5,000th Robot and Discloses Technology Roadmap

This page provides a summary and source link only.