Most humanoid startups begin with a dream of the home and end up pitching factories. Fourier Intelligence (傅利叶智能) inverted the path: it spent a decade building rehabilitation robots for stroke and mobility patients, then grew a humanoid out of that clinical muscle. Its bet is that the hardest, most valuable job for a humanoid is not welding cars but helping people recover and age.
That is a quieter thesis than the headlines about warehouse bots, and arguably a harder one to fake.
Key takeaways
- Fourier started in 2015 as a rehabilitation-robotics company (upper-limb, gait, and passive-motion trainers) before entering humanoids.
- The GR-2, launched in 2024, carries up to 53 DoF, weighs about 63 kg, stands 1.75 m, and uses 12-DoF dexterous hands with six array-type tactile sensors.
- Its latest E-round financing totaled close to 800 million yuan (≈ US$110 million) per company announcements reported by state-affiliated media.
- Fourier says its robots support over 3,000 institutions and hospitals across more than 40 countries.
A rehab company that learned to walk
Fourier's origin story is unusual. Long before "embodied intelligence (具身智能)" became a buzzword, the Shanghai firm was shipping medically regulated devices: the ArmMotus upper-limb trainer, gait-analysis and training systems, and passive motion equipment for limb rehabilitation. That clinical background shaped how it thinks about humanoids — as tools that must be safe around fragile bodies, not just strong on a factory floor.
The GR-1, revealed in 2023, was Fourier's first full-size humanoid and, the company says, entered global mass production and delivery. The GR-2 followed in 2024 with a clear "Made for AI" framing: a platform for developers to teach embodied agents. Technically it is serious hardware — 53 DoF, peak joint torque above 380 N·m from its self-developed FSA 2.0 actuators, a 1,100 Wh battery, and roughly 2 hours of runtime.
The caregiving bridge
Where Fourier diverges from peers is the scenario. Rather than lead with automotive assembly, it built what it calls an "embodied-intelligence rehabilitation port (具身智能康复港)" centered on the GRx series, with five interaction modules: triage consultation, cognitive training, upper-limb rehab, motor-function重建, and remote rehab. The Care-bot GR-3, a full-size humanoid aimed at companionship, is designed for roles like keeping company with elderly people living alone or engaging children.
In a pilot "humanoid + rehabilitation" setting built with a Shanghai medical center, a GR-1 greeted visitors at a reception desk while therapists tele-operated a GR-2 to run activities-of-daily-living training — stacking blocks, moving cups — for patients. The point is not autonomy today; it is a verifiable technical path from tele-operation to assisted therapy.
Why rehabilitation is a strategic moat
Fourier argues that humanoid robots and therapists are complementary, not competitive. The pitch rests on a real problem: shortages of rehabilitation professionals leave treatment gaps, especially for aging populations. A robot that can deliver repetitive, monitored, data-rich sessions — and free clinicians for judgment-heavy work — fits a need that exists now, not in some distant consumer future.
The company has also pushed openness. Alongside GR-1 and GR-2, Fourier released N1, its first open-source humanoid, and its developer toolkit supports NVIDIA Isaac Lab, ROS, and Mujoco. That echoes its belief that a caregiving ecosystem needs third-party developers building therapy and assistance applications.
The funding signal
Fourier's latest E-round financing, totaling close to 800 million yuan (≈ US$110 million), drew investors including state-linked Shanghai funds and regional venture firms, according to company announcements carried by state-affiliated media. For a firm rooted in medical devices, that capital is a vote of confidence in the rehabilitation-and-caregiving roadmap rather than in another general-purpose humanoid.
Honest caveats
The caregiving vision is still largely pilot-stage. Tele-operation, not autonomy, does the therapeutic work today, and clinical adoption will hinge on regulatory clearance, liability, and demonstrated outcomes — slow, conservative gates unlike the flashy factory demos elsewhere. A humanoid that touches patients must clear a far higher safety and trust bar than one that moves parts bins.
What readers can do now
- Clinicians and hospital operators should watch Fourier's rehab-port pilots for published outcomes data before assuming cost savings; ask for measured therapy-hours gained, not demos.
- Robotics developers interested in non-industrial humanoids should evaluate the open-source N1 and Fourier's toolkit as a safer, care-oriented entry point than factory-first platforms.
- Investors tracking the sector can use Fourier's E-round and its rehabilitation revenue base as a contrast case to hardware-only humanoid startups with no recurring clinical income.
Honest limitations
This piece relies on Fourier's official GR-2 specification sheet, the company's own event reporting (GReAT 2025 summit), a China Science and Technology Association (科创中国) outcome entry on the rehab port, and state-affiliated media (China Economic Net) for the financing figure. The "close to 800 million yuan E-round" number is reported, not independently audited here, and I could not verify per-country institution counts beyond Fourier's own stated "3,000+ institutions, 40+ countries." I have not visited a Fourier deployment or handled a GR-2. My perspective is that of an external observer of China's robotics and rehabilitation sector, not a Fourier stakeholder.