China/spotlight
OriginFlow reads muscle signals to teach robots to use their hands
Qin Shentao registered the company in his third year of a Tsinghua doctorate. Within nine months it had raised over 500 million yuan.
Humanoid robots can now walk and wave. What they cannot yet do well is fold a shirt, and the reason is usually data. Teaching a robot hand to be gentle requires recordings of human hands being gentle, with force and touch included, and few such recordings exist. OriginFlow, founded by a Tsinghua doctoral student, is building a way to collect them.
The founder
Qin Shentao was born in 2001 in a mountain village in Jincheng, Shanxi, according to a profile by Chinese outlet Zimubang, republished by 36Kr in June 2026. He entered the Harbin Institute of Technology in 2019 to study mechatronic engineering and, by that account, ranked first in his major. He worked on force perception for legged robots and led a student team in the university’s entrepreneurship incubation park.
In his final undergraduate year he joined the Shenzhen Institute of Technology and Innovation, founded by Professor Li Zexiang, and Lu Qi’s MiraclePlus programme. He had a rough idea for a neural-interface base model but judged the timing and the use case wrong. He then began a doctorate at Tsinghua’s School of Vehicle and Mobility. In August 2025, in his third doctoral year, he registered OriginFlow, about a month after Meta published research on a non-invasive neural motor interface, Zimubang reports. Zimubang quotes Cao Xi, founding partner of Monolith Capital, calling him “a natural entrepreneur”.
The product
OriginFlow’s NeuroScale system uses surface electromyography: sensors on the skin pick up the electrical activity of muscles, and a proprietary model translates the signals into continuous hand postures, force values and tactile feedback, according to Gasgoo. The first device is a wristband. The company argues that this beats first-person video, teleoperation and motion-capture gloves, which it describes as limited in scale or commercial readiness. Gasgoo’s phrase for NeuroScale is “a new data foundation for robots to achieve fine, anthropomorphic dexterous manipulation”. That is the company’s claim, not an independent test.
Two applications are named. In industrial manufacturing, OriginFlow says it is working with leading global manufacturers on data collection for precise, repetitive and hazardous processes. At home, it is working with 58 Group to record laundry folding, cleaning and kitchen tasks, building a skill database for housekeeping robots.
The money
Gasgoo, dated 22 May 2026, reports more than 500 million yuan raised over several months. The angel round was co-led by BlueRun Ventures and Oasis Capital. A strategic round added 58 Strategic Investment, Matrix Partners China and the Tsinghua University Seed Alumni Fund. A Pre-A1 round was led exclusively by Monolith Capital, with Matrix increasing its stake. Zimubang counts ten investors and three rounds between December 2025 and May 2026.
VnExpress, in October 2026, lists Qin, aged 25, among a group of Tsinghua doctoral students who started physical-AI companies and drew large rounds.
What is unproven
Neither article reports revenue, shipments or accuracy figures. The case for OriginFlow rests on a thesis: that robots will be limited less by their bodies than by the data they learn from, and that muscle signals are a cheaper way to gather it. Investors have paid for the thesis. The next evidence will be whether the wristband’s recordings produce robots that handle objects better.
