Fourier connects humanoid interaction, mobile manipulation and rehabilitation

Fourier develops robots for human interaction, physical assistance and rehabilitation. That breadth makes the company relevant to embodied AI, but it also makes product selection important. A social humanoid, a mobile manipulation platform and a specialised rehabilitation system solve different problems, with different evidence and operational responsibilities behind each purchase.

01 / ProductThree product paths share an engineering foundation

The Fourier company overview traces the company from rehabilitation robotics into humanoid systems. Its current portfolio includes GRx humanoids, rehabilitation platforms and open-source resources. This creates a useful way to understand Fourier: the company develops bodies, actuation and software for physical interaction, then packages those capabilities differently for research, services and clinical environments.

The GR-3 series page distinguishes GR-3, positioned around interactive companionship, from GR-3C, which adds a lidar module and a more functional exterior. GR-3 combines vision, audio and touch with expressive behavior. These are mechanisms for interaction; phrases about emotional connection describe the vendor’s product intent rather than evidence that the robot understands human feelings or can provide independent care.

The GRW page describes a mobile dual-arm robot for lifting, transport and object handling. It lists interchangeable end effectors and FOCUS fleet management. A wheeled platform shifts the engineering emphasis away from bipedal balance and toward navigation, reach and manipulation. That can be a sensible distinction when the useful job takes place on accessible floors.

RehabHub brings together specialised rehabilitation equipment and a connected software environment. It should be evaluated as its own product category. A hospital buying a rehabilitation system and a laboratory buying a humanoid may be engaging the same company, but they require different operating evidence, training and acceptance criteria.

02 / AudienceThe best fit depends on the human interaction you need

A human-robot interaction laboratory can consider GR-3 when its research requires a physical social presence rather than a screen-based interface. The research question might concern how people respond to gestures, touch cues or turn taking. Choose an experiment in which the body’s contribution can be observed separately from the language system. Otherwise an expressive demonstration can hide uncertainty about which component produced the effect.

A facilities or logistics team may find GRW more relevant when it needs movement between stations and controlled handling at each stop. First describe the objects, reach heights and available floor space. A mobile robot must be able to approach the work in a useful orientation, not merely fit through the doorway. Small layout changes can materially alter the handling problem.

A clinical team should begin with the intended rehabilitation workflow, staff supervision and exact device configuration. Fourier’s rehabilitation catalogue is evidence of a commercial product line, not a universal claim about patient outcomes. Clinical adoption decisions require the relevant device documentation and evidence for the intended setting rather than conclusions borrowed from humanoid marketing.

For developers, Fourier Nexus presents hardware resources, open-source algorithms and data sharing, including N1 and ActionNet. It also describes some future technology releases as planned. Distinguish resources actually available now from the ecosystem roadmap before scheduling a research programme around them.

03 / WorkflowA proposed GR-3 trial separates dialogue from motion

Imagine a supervised reception experiment in which a robot greets a visitor and indicates a staffed desk. This is a proposed evaluation, not a tested Fourier installation. Start with a small set of approved interactions and an operator nearby. Define when the robot should remain still, when it may gesture and when it must hand the conversation to a person.

Use the current GR-3 developer overview to map the experiment onto the delivered hardware. Its detailed table distinguishes base and optional dexterous hands, describes joint motion limits and identifies communication interfaces. The lesson is to choose the actual configuration needed for the gestures. An upper-bound joint count should never be assumed to describe the least expensive or default package.

Build the trial in stages. First run the conversation with the robot stationary and observe whether visitors understand the instructions. Then introduce approved head and arm movements. Finally test small variations in visitor position and background noise. Each stage answers a different question and makes a failure easier to diagnose.

Keep the language boundary explicit. If the visitor asks for information outside the approved knowledge, the robot should redirect to the staffed desk instead of improvising an operational commitment. The research log should distinguish a speech recognition problem, an inappropriate answer and a motion issue. Treating every failed interaction as a single AI error would make the next iteration less informative.

At the end, compare against a conventional sign or staffed greeting under the same conditions. Measure successful routing, repeated questions, operator interventions and visitor willingness to engage. Do not infer improved wellbeing or care outcomes from a pleasant interaction. A useful result may simply establish that a particular gesture makes directions easier to follow.

04 / PricingFourier prices are tied to the product and installation

Fourier routes enquiries through its business contact page; the GRW and RehabHub pages also direct readers to sales. The reviewed pages did not establish a public standard price for these configurations as consulted on 3 October 2026. Request a quote for a named product and purpose rather than assuming that the different families share one commercial package.

For GR-3, make hands, computing, developer access and operator training explicit. For GRW, include the end effector and task software required for the chosen objects. For a rehabilitation installation, establish the devices, software connections and service provision around the intended clinical workflow. A robot body price alone does not describe those scopes.

Also distinguish open resources from commercial support. Access to a public dataset or hardware resource does not establish what assistance is included with a purchased robot. Ask which software versions are supported, how configuration changes are handled and whether the experiment can remain reproducible after future updates.

05 / DistinctionsEmbodiment changes what an interaction can achieve

Fourier’s range makes a distinction visible that software-only AI often hides: the physical interface matters. Touch sensing, posture and a moving arm create interaction possibilities that a chatbot does not have. They also introduce timing, reach and supervision requirements. The useful design question is which physical behavior improves the task enough to justify that additional complexity.

The Unitree blueprint provides a comparison for teams choosing robotics hardware as a development platform. Compare access to control, sensors and supported editions. A laboratory should select the body that enables its experiment rather than the robot with the most broadly worded capability description.

The 1X blueprint offers context for humanoid assistance in human spaces. Fourier’s social interaction and rehabilitation roots suggest another starting point: a carefully bounded interaction or assistive workflow. Neither company’s broad mission should be read as proof of unsupervised competence across every household or care task.

06 / QuestionsResolve configuration discrepancies before designing the experiment

The current GR-3 marketing page and developer overview do not align on every specification, including walking speed and camera description. This article avoids choosing a preferred figure from those differences. Obtain the specification sheet for the exact delivered revision before relying on a sensor modality or movement limit in the research design.

The hand distinction is equally important. The documentation ties its maximum actuator count to an optional hand configuration. If an experiment depends on fine finger movements or tactile feedback at the hand, confirm those capabilities directly rather than assuming that all GR-3 units have identical end effectors.

For GRW, fleet coordination and interchangeable tools are promising integration concepts, but the decisive evidence is the chosen task with the chosen tool. A demonstration of moving one object does not establish handling across an entire store or ward. Introduce variation deliberately and preserve a clear route for a person to resolve exceptions.

07 / DecisionPick the product around a measurable human task

Fourier is most useful to evaluate when the team can explain why a physical robot belongs in its workflow. That may be a social research question, a mobile handling need or a specialised rehabilitation application. Keeping those intentions separate makes the first conversation with the supplier much more productive.