A night shift in a Huzhou, Zhejiang warehouse runs with no human supervisor on the floor. A full-size humanoid named STAR1 moves parcels, scans barcodes, and sorts them — repeating tasks it learned from a handful of real-world demonstrations rather than a hand-written script.
That robot comes from Noetix Robotics (星动纪元), a company that did not exist two years ago and is now one of the most funded humanoid (人形机器人) startups in China. Its rise says less about one clever machine and more about how fast the country's embodied-AI (具身智能) supply chain has matured.
From a university lab to a factory floor
Noetix was founded in August 2023 and grew out of the Institute for Interdisciplinary Information Sciences at Tsinghua University (清华大学交叉信息研究院). Its founder, Chen Jianyu (陈建宇), is a Tsinghua assistant professor who did his PhD at UC Berkeley under Masayoshi Tomizuka, a pioneer of mechanical control theory. The company describes itself as the only Tsinghua-held humanoid robotics venture, and it has kept research and hardware under one roof.
That matters because humanoids are usually built from parts sourced across a fragmented supply chain. Noetix instead designs its own joint modules, control units, and motors, then scales up to full bipeds, wheeled bodies, and dexterous hands. The bet is that owning the "body" as tightly as the "brain" is what lets a startup ship at volume.
What STAR1 actually does
STAR1 is a full-size bipedal humanoid with 55 degrees of freedom. In outdoor tests it has run at 3.6 meters per second, and its upper body pairs a 7-degree-of-freedom arm with a 12-degree-of-freedom dexterous hand called XHAND1. Those numbers put it in the same conversation as the better-known Chinese humanoids, but Noetix's pitch is breadth: the same software stack also drives Q5, a wheeled service humanoid with 44 degrees of freedom built for malls and retail floors.
In industrial pilots, STAR1 has been trained at a factory run by the Beijing Machinery Automation Research Institute (北自所) to carry, sort, and scan packages. In service settings, Q5 has been tested with partners including Haier (海尔) and Century Golden Resources (世纪金源) for greeting, guiding, and carrying items in stores.
The model behind the machine
The differentiators Noetix talks about most are software. In December 2024 it released ERA-42, an end-to-end VLA (vision-language-action) model that folds visual understanding, prediction, and action into one network — the company says it is the first embodied model able to control both the full body and the dexterous hand. In May 2025 it open-sourced VPP (Video Prediction Policy), a generative model that learns from video of many robot shapes and needs only a little real-world data to fine-tune.
The claim, if it holds, is economically large: real robot demonstration data is the most expensive input in the field. A model that learns mostly from video could collapse the cost of teaching a new task.
Why the money followed
Noetix's commercial footprint is the part outside observers can partly verify. The company says it delivered more than 200 units in 2025, with overseas customers making up roughly half of its orders, and that cumulative orders have passed RMB 500M (≈ US$70M / HK$550M). It also states that nine of the world's ten largest technology firms by market cap have tried its robots — a figure we treat as the company's own and have not independently confirmed.
Investors have been explicit. In July 2025 Noetix closed a Series A of about RMB 500M (近5亿元, ≈ US$70M / HK$550M) led by CDH VGC (鼎晖VGC) and Haier Capital (海尔资本). In November 2025 it raised an A+ round of about RMB 1B (近10亿元, ≈ US$140M / HK$1.1B) led by Geely Capital (吉利资本), with BAIC (北汽产投) as a strategic backer. Then in March 2026 it announced a strategic round of RMB 1B (10亿元) that pushed its valuation above RMB 10B (≈ US$1.4B / HK$11B), with new investors including Samsung, Singtel (新加坡电信), and Porsche-backed CICC, bringing the total number of backers to 16 across tech, automotive, logistics, and semiconductor industries.
The March 2026 round was reported by China Securities Journal (中国证券网), Guangming Network (光明网), and the Beijing Municipal Economy and Information Technology body, which also named Noetix Beijing's third humanoid unicorn after Galbot and Robotera.
What readers can do now
- Watch the "video-trained" claim, not the demo videos. If Noetix or a peer can show a new task learned from mostly video with minutes of real data, that is the real inflection — far more than a polished walking clip.
- Track overseas revenue share. Noetix says half its orders are foreign. Sustained export share is a cleaner signal of product maturity than funding headlines.
- Don't read a valuation as a verdict. A >RMB 10B paper valuation reflects investor belief in 2026, not delivered revenue. Treat it as a bet, not a scoreboard.
Honest limitations
This article relies on company announcements and Chinese financial and technology media; we did not independently audit Noetix's delivery or order figures. The "nine of ten" customer claim and the specific task-efficiency numbers are company statements, not third-party-verified data. The funding chronology — Series A, A+, and the March 2026 strategic round — is consistent across multiple Chinese outlets, but private-round terms can shift after announcement. We have not tested STAR1 ourselves, and humanoid reliability in unstructured real-world shifts remains unproven at scale.
Key takeaways
- Noetix (星动纪元), a 2023 Tsinghua spin-off, shipped 200+ humanoids in 2025 and reached a valuation above RMB 10B after a March 2026 strategic round.
- Its core bet is full-stack control of both body and brain, plus a VLA model (ERA-42) and an open-source video-prediction policy (VPP) meant to cut reliance on costly real-robot data.
- The story is less about one robot and more about how fast China's embodied-AI supply chain can turn a lab spin-off into a funded, exporting humanoid maker.