Fourier GR-3: Can a Humanoid Be a Companion?

That is a very different promise from "this robot can move boxes." It also needs a more careful reading. A robot can be emotionally responsive without being a safe, affordable, autonomous home-care product.

The useful way to think about GR-3 is not "humanoid companion robot: solved." It is this: Fourier is testing whether a full-size humanoid can borrow the comfort cues of companion robots while keeping the reach, walking, teleoperation, and manipulation hardware of a service robot. That puts GR-3 in a narrow middle lane between LOVOT, ElliQ 3, 1X NEO, and Hello Robot Stretch 4.

What is the Fourier GR-3?

Fourier GR-3 is the third-generation member of Fourier's GRx humanoid line and the first one the company explicitly frames around care, companionship, and human-centered service. In the ui44 database it sits in the Humanoid category with Active status, no public price, and a release date of August 2025.

The core public specs are concrete enough to compare. These are manufacturer figures, checked on October 14, 2026 against the GR-3 Series product page and launch announcement:

The current GR-3 brochure describes whole-body teleoperation, an upgraded backend UI, and algorithm development on a client-server architecture. The August 2025 launch also announced plans for interaction APIs. The brochure now describes stereo vision and an AI depth camera; the table preserves the launch's RGB/structured-light description, so buyers should confirm the sensor configuration being offered. That tells us a lot about the real market: this is still a platform for labs, integrators, care pilots, public service demos, and enterprise research teams. It is not a boxed consumer robot you buy for a parent this weekend.

The shift is in the product language. Fourier GR-2 was mostly a research and enterprise humanoid. Its official brochure lists 175 cm, 63 kg, up to 53 DoF, standard 6-DoF hands with optional 12-DoF hands, about 2 hours of battery life, and ROS/Isaac Lab/MuJoCo support; it does not list a public price. GR-3 keeps the humanoid-platform DNA but adds a softer shell, more social interfaces, and a care-first story.

Why does Fourier call GR-3 a Care-bot?

Fourier's care claim has three layers.

First, the body is intentionally less industrial. The official materials describe warm neutral tones, cushioned surfaces, and upholstery-like coverings. That may sound cosmetic, but physical design matters when a robot is 71 kg and moving near children, older adults, or clinical patients. A softer-looking robot is not automatically safe, but it is less intimidating than a bare metal lab platform.

Second, GR-3 has a multimodal interaction stack. Fourier says it combines vision, audio, and touch into a real-time emotional processing engine. In practical terms, that means the robot can orient toward a speaker, track a face, respond to touch, blink, move its eyes, and trigger gestures that look more social than a standard utility robot.

Third, GR-3 uses what Fourier describes as a dual-path response architecture: fast rule-based reactions for reflexive behavior, plus slower LLM-based contextual dialogue for more complex conversation. That split is important. A care robot cannot wait for a cloud-style chat response before reacting to touch, balance, or proximity. It needs fast local behaviors for physical interaction and slower reasoning for conversation.

The limitation is equally important: none of this proves eldercare autonomy. The official language says GR-3 can act as a service assistant or responsive companion in public settings, while assistive care applications such as mobility support, health monitoring, and rehabilitation are described as future or potential applications. That wording is worth preserving. GR-3 is care-centric in positioning and interface design, not a verified replacement for a caregiver.

How is GR-3 different from companion robots like ElliQ and LOVOT?

Most companion robots avoid the hardest physical problems. ElliQ 3 is a stationary AI companion for older adults. Its feature descriptions cover proactive conversation, medication reminders, wellness programs, video calling, photo and message sharing, games, and social connection. Its US membership pricing, checked October 14, is a $249 one-time lease initiation fee plus $59 monthly, $588 prepaid annually, or $936 prepaid for 24 months. The longer plans work out to $49 and $39 per month, respectively; those are equivalents, not monthly billing options.

LOVOT goes the other direction: it is mobile, warm, sensor-rich, and emotionally expressive, but it is not trying to manipulate the world. In Japan, the basic ちゃ LOVOT 3.0 example costs ¥577,500 for orders through October 25, 2026, including tax, administration and initial shipping. Standard monthly care starts at ¥9,900/month; an upfront software-use option is also listed, so monthly billing is not universal. The announced basic-color prices are ¥599,500 from October 26 and a tentative ¥649,000 around mid-January 2027, with the final January price and exact date still pending. Other colors cost more.

LOVOT's founder described over 50 sensors in 2019. The current 3.0 has a 43 cm held height and 4.6 kg weight excluding clothes, touch response, room mapping, person recognition, and a charging nest. It exists to be loved, not to lift a cup or support a person standing up.

Casio Moflin and Mirumi make the same trade-off in smaller forms: tiny, low-pressure companionship over physical utility. Casio describes Moflin as a portable emotional companion, while Mirumi's maker describes a charm robot that turns toward sound and touch. Familiar, targeting 2027 according to an original founder interview, is a dog-sized quadruped built around body language and emotional presence, without a screen or spoken language. Its manufacturer describes facial expressions and vocalizations, with data stored on the device and owners controlling cloud sharing. That is a storage and sharing statement, not proof that all processing is local. The reported launch target is not a delivery guarantee.

GR-3 is trying to merge some of that social-emotional design language with a full-size humanoid body. The upside is obvious: it has hands, legs, height, and the potential to interact with human-scale spaces. The downside is just as obvious: every kilogram, joint, and moving limb raises the safety and reliability bar.

Scroll sideways to inspect the full chart.

That is why GR-3 should not be judged by the same standard as a tabletop companion. A robot that chats from a desk mostly has to be engaging, private, reliable, and easy to maintain. A full-size care humanoid has to be all of that while also proving it can move safely around people and recover from physical failures.

Is GR-3 closer to 1X NEO or Hello Robot Stretch 4?

For home buyers, the more useful comparison is with robots that can physically help.

1X NEO is the most obvious consumer-facing contrast. The manufacturer's order page lists 5 ft 6 in (about 168 cm), 66 lb (about 30 kg), a $20,000 Early Access ownership price, and a $200 reservation deposit. It advertises 4 hours of runtime, a soft lattice body, stereo fisheye cameras, four beamforming microphones, and a home-first chore pitch. 1X also markets scheduled Expert Mode, where a human operator can guide chores the robot cannot yet do by itself.

That is a very different bet from GR-3. NEO is lighter, explicitly home-focused, and framed around household chores. GR-3 is heavier, less transparent on price, and framed around care, public service, clinical use, research, and eventually personal spaces. NEO's biggest question is whether home chore autonomy arrives quickly enough. GR-3's biggest question is whether a care-centric full-size humanoid can be accepted in real human spaces before it is priced or packaged for private homes.

Hello Robot Stretch 4 is the cleaner assistive comparison. Stretch 4 is not humanoid. Its purchase page lists $29,950 for the robot and $1,495 for a separate charging dock; certification is for laboratory/research use. It is not generally sold for individual use. That does not erase its home work: Stretch Assist describes selective remote testing and in-home co-design pilots with engineers and an occupational therapist.

The product specifications retain the practical advantages: an omnidirectional wheeled base, a telescoping arm, 2.5 kg extended / 4 kg retracted payload, self-charging with the dock, 8 hours under light CPU load, open ROS 2/Python support, and a compact 45 cm diameter footprint on the HTML page. Buyers planning tight clearances should confirm dimensions: its linked technical PDF gives a conflicting 43 cm.

Stretch 4 is less emotionally expressive than GR-3, but its design is focused: mobile manipulation in real homes. For assistive tasks, that focus can be more valuable than a human-shaped body. A wheeled base is easier to stabilize. A single arm is easier to supervise. A robot that can reach shelves, pick objects, and dock itself may be more useful than a humanoid that looks caring but still needs careful teleoperation.

What does GR-3 prove for home robotics?

GR-3 proves that the companion-robot category is expanding upward. It is no longer just small pets, tabletop displays, and rolling smart-home cameras. The same design goals are moving into full-size humanoids:

• a body that feels less threatening
• touch response instead of touch avoidance only
• eye contact, blinking, and social timing
• conversation that can adapt to context
• teleoperation and APIs for supervised real-world use
• a hardware path toward actual assistance, not just companionship

That is meaningful. Home robots have a trust problem. People do not only ask "Can it do the task?" They ask whether they can relax around it. A care robot that moves through eldercare, rehab, or family spaces needs emotional legibility: people should understand when it sees them, hears them, is about to move, or is asking for help.

But GR-3 does not yet prove that a full-size companion humanoid is the best home robot shape. Its current facts point to a platform stage:

Fourier's earlier hardware does have research evidence. The GR00T N1 paper reports deployment on GR-1 for language-conditioned bimanual manipulation. That gives the platform lineage more substance than a concept render, but it does not establish GR-3's safety or effectiveness in care and personal environments.

What should buyers ask before trusting a care humanoid?

If GR-3 or a similar care-centric humanoid appears in a pilot, showroom, or future preorder flow, the right questions are practical:

1. What tasks are autonomous today? Conversation, greeting, and gesture are not the same as safe transfer assistance, object retrieval, or health monitoring.
2. What happens when it is uncertain? A care robot needs graceful refusal, handoff, or teleoperation, not improvisation near a vulnerable person.
3. Where does perception data go? GR-3's pitch depends on face tracking, voice input, touch sensing, and LLM dialogue. Buyers should ask what is processed locally, what is uploaded, and how long data is retained.
4. What safety evidence exists beyond soft materials? Cushioned surfaces and pressure sensors are helpful design choices, but they are not the same as an independently validated care environment.
5. Who maintains it? A 71 kg humanoid with up to 55 DoF is not a gadget. Homes and care facilities need service contracts, spare parts, remote diagnostics, and clear downtime expectations.

These questions are not meant to dismiss GR-3. They are the difference between a credible robotics pilot and a marketing demo.

Should you care about Fourier GR-3 now?

Yes, if you follow the future of home and care robots. GR-3 is a useful signal that humanoid makers are learning from companion robots: touch, expression, softness, social timing, and emotional context are becoming part of the hardware roadmap, not just app-layer decoration.

No, if you are shopping for a practical home-care robot today. For an older adult who needs daily social support, ElliQ 3 is far more available and purpose-built. For emotional companionship, LOVOT, Moflin, Mirumi, or future Familiar-style robots are closer to real consumer products. For assistive manipulation research, Stretch 4 is more transparent on price and deployment status. For a humanoid home-helper preorder, 1X NEO is the clearer consumer-facing bet, even with its own autonomy and operator-handoff caveats.

Fourier GR-3 matters because it asks the right product question: can a humanoid be accepted as a caring presence, not just a worker? The answer is not proven yet. But the question itself is important, and it is exactly where the next phase of home robotics is heading.