Picture a Shanghai rehabilitation clinic where a stroke patient reaches toward a cup held by a robotic arm. The arm does not just move; it senses the patient's effort through its joints and adjusts resistance in real time. The robot is a Fourier (傅利叶智能) device — and the same force-control instinct that lets it assist a recovering hand is now inside the company's humanoid (人形机器人).
Fourier is the rare Chinese robotics company that did not start with humanoids. For nearly a decade it built rehabilitation robots and exoskeletons (外骨骼) for hospitals. That history shapes how it approaches the humanoid race — and why its latest machine, the GR-3, is aimed at care rather than factories.
A rehabilitation company first
Founded in 2015 in Shanghai by Gu Jie (顾捷), a Shanghai Jiao Tong University graduate, Fourier began with therapy devices. In 2017 it released what it describes as China's first commercial lower-limb exoskeleton, the X1, letting patients with mobility impairment stand and walk in training. Over the following years it deployed upper- and lower-limb rehab systems to more than 2,000 institutions and hospitals across 40-plus countries.
That rehab business did two things a pure humanoid startup could not: it generated revenue and, crucially, taught Fourier to control force. Assisting a fragile human limb demands precise, compliant torque — the opposite of a rigid industrial arm. When the company started a general-purpose humanoid project in 2019, it carried that control stack with it into the broader embodied intelligence (具身智能) push.
Fourier's rehab catalog is broader than its humanoids suggest. Its ExoMotus M4 lower-limb exoskeleton helps patients with mobility impairment stand and walk in early recovery, using self-developed integrated actuators for compliant gait training. At WAIC 2025 the company unveiled an "embodied intelligence rehab port" (具身智能康复港) where GR-series humanoids act as guides and upper-limb trainers, blending the old rehab business with the new humanoid one. That continuity — selling into the same hospitals for a decade — is the asset rivals built from scratch simply do not have.
From GR-1 to GR-3
Fourier unveiled its first general humanoid, the GR-1, in 2023 and began mass production. The GR-2 followed in September 2024: about 175 cm tall, 63 kg, up to 53 degrees of freedom, with an optional 12-degree-of-freedom dexterous hand and six array-type tactile sensors. Its self-developed FSA (Fourier Servo Actuator) 2.0 reaches peak torque above 380 N·m, and the machine runs roughly two hours per charge.
The GR-3, shown in 2025 and debuted abroad at CES in early 2026, is the clearest statement of intent. At 165 cm and 71 kg with 55 degrees of freedom, it adds full-body tactile sensing, multimodal emotional interaction, and dual hot-swappable batteries giving about three hours of runtime. Fourier prices it at 115,000 yuan (≈ US$16,000) as of early 2026. The positioning is explicit: a "carebot" for eldercare and rehabilitation, not a factory handler.
Why care, not cargo
Fourier's differentiation is interaction. A rehab or care robot spends its time close to people — reading intention, moderating its own strength, surviving unpredictable contact. The company argues that its exoskeleton heritage gives it an edge in exactly these force-sensitive, human-centered tasks, where a Unitree or an AgiBot built for speed and manipulation may be less at home.
It has also embraced openness as a strategy. Fourier released an open-source humanoid, the Fourier N1 (about 1.3 m, 38 kg, 23 degrees of freedom), publishing drawings, code and assembly guides, and open-sourced a robot-action dataset, Fourier ActionNet, to seed a developer community.
The business underneath
Fourier remains private. The funding predates the GR-3's early-2026 CES debut: in January 2025 it closed a Series E round of about 800 million yuan (≈ US$111 million) with investors including Prosperity7 and Shanghai state-backed funds. The company says it shipped more than 300 humanoids in 2025 and has delivered over 10,000 robots cumulatively — the vast majority rehabilitation machines, not humanoids. Rehab revenue, in other words, still funds the humanoid dream.
The cautious timeline
Gu has been plain about the gap between demo and deployment. He expects humanoids to appear first in large rehabilitation centers, rehab hospitals and community clinics within two to three years, and only to reach homes in five to ten. That is a slower, more grounded horizon than rivals promising factory takeovers next quarter. The restraint is a competitive choice as much as a technical one: by aiming at care, Fourier avoids a head-to-head price war with factory-focused rivals while leaning on revenue its rehab line already earns.
Honest limitations
Figures here draw on Fourier's official website, Chinese state media (China.com) and robotics trade profiles. Shipment and delivery counts (300+ humanoids in 2025; 10,000+ cumulative) are company-stated and not independently audited. The GR-3's care capabilities are demonstrated in controlled rehab scenarios, not yet validated across diverse home environments. I have not tested any Fourier machine and cannot vouch for real-world reliability, safety certification status, or how its force control compares head-to-head with peers. Currency conversions are approximate.
What readers can do now
- If you evaluate care robots, prioritize force-compliance and safety certification over flashy manipulation demos.
- For rehab buyers, ask whether a humanoid adds capability a dedicated rehab device lacks — often it does not yet.
- Track Fourier's regulatory clearances in healthcare; eldercare adoption will hinge on certification, not capability videos.
