Fourier Intelligence Market Research

1. Introduction

Fourier Intelligence's differentiating thread within this group is "differentiation through rehabilitation-based healthcare": it is the only vendor among the ten that started with rehabilitation robots and treats medical eldercare as its primary scenario. While the other vendors compete over factory and logistics scenarios, Fourier advocates "how robots can serve people," turning its humanoid robot into a Care-bot for the elderly, patients, and families. Its GR-3's soft-shell design, facial-expression interaction, and dual-route cognition architecture form a stark contrast with industrial humanoids' metal bodies and task-first approach.

1.1. Company Overview

1.2. Development History

1.3. Business Model & Shipment Scope

Fourier's commercial foundation rests on two wheels: rehabilitation medical devices + humanoid robots: its rehabilitation business (Intelligent Rehab Port, RehabHub) has entered medical institutions in multiple countries and regions (per official statements, without specific figures), providing stable cash flow; its humanoid robots advance through whole-unit sales + scenario solutions (eldercare institutions, high-end service centers, exhibitions). No authoritative figures are available for shipment volume, historical GR-1 / GR-2 sales, or rehabilitation revenue, all marked [To be filled]. GR-3 pricing has not been disclosed and is marked [To be filled].

2. Glossary

3. Functional Description & Product Lines

3.1. Product Line Overview

3.2. Core Capabilities of the GR-3 Care-bot

1. Safe human-robot contact: soft-shell design (automotive-grade fabric + eco-friendly foam) + 31 distributed pressure sensors across the body, aimed at scenarios involving direct physical contact with the elderly and patients — a design constraint industrial humanoids do not face.
2. Multimodal perception: a 4-microphone array + RGB + structured-light camera supporting speech, vision, and proximity perception.
3. Emotional interaction: the LED expression face provides non-verbal feedback; at WAIC 2026 it demoed recognizing vague commands like "I'm thirsty" and responding.
4. Continuous availability: ~3-hour average endurance + hot-swappable battery support round-the-clock duty in eldercare institutions.
5. Fine manipulation: a 12-DOF dexterous hand covering everyday care actions such as handing over objects and opening doors.

4. Platform Architecture

4.1. Dual-Route Cognition Architecture

GR-3 uses a dual-route cognition architecture: the rule-controller fast system handles low-latency reflexive reactions and the safety bottom line (e.g., immediate stop when contact pressure exceeds limits), while the LLM slow system handles semantic understanding, vague-command parsing, and task planning. The two run in parallel with the fast route as fallback — a classic engineering solution to the "low-latency safety + high-level intelligence" tension in embodied-interaction scenarios.

4.2. Rehabilitation Business Foundation

Fourier's platform uniqueness lies in its ten years of medical-grade delivery experience: the compliance experience of getting rehab devices into medical institutions in multiple countries, clinical-collaboration processes, and after-sales service networks are directly reused as the delivery foundation of the Care-bot. Compared with industrial humanoid vendors starting from scratch, Fourier's channels and compliance assets are its differentiation moat.

4.3. GR-3 Dual-Route Cognition Architecture Diagram

Fig. 5-1 | GR-3 dual-route cognition architecture (fast and slow systems in parallel)

Note: drawn based on this article's analysis of the dual-route cognition architecture; parameters such as sensor counts come from GR-3's official release statements.

5. Harness Design

5.1. L1 Context Engineering Layer

GR-3's multimodal perception (speech + vision + contact pressure) forms its context input stack; in the dual-route cognition architecture, the slow system parses vague commands like "I'm thirsty" into executable task intents. The special requirement of medical scenarios — continuous understanding of the patient's condition — constitutes the domain-specific context of L1, but the specific engineering components (retrieval, memory retrieval, command routing) are not publicly disclosed.

5.2. L2 Tools & Execution Layer

The 12-DOF dexterous hand + soft-shell safety body form the execution layer. Fourier's unique constraint in the execution layer is sharing physical space with humans: the 31 distributed pressure sensors attach contact-safety detection to every execution action, and the fast system's rule-based control runs directly on L2 for ultra-low-latency safety arbitration. The hot-swappable battery supports 24/7 standing-duty execution.

5.3. L3 Orchestration & Control Layer

The dual-route architecture is naturally a layered implementation of L3: the fast system handles reflex-level control while the slow system (LLM) handles task-level planning. Multi-stage home-scenario tasks (guest reception → object delivery → home-security series) are orchestrated by the slow system. No explicit workflow engine or multi-robot coordination is disclosed.

5.4. L4 Memory & State Layer

The Care-bot's core differentiation need is to remember people: recognizing regular visitors' faces and remember preferences and medical-history context. Official demos include face recognition and personalized interaction, pointing to the necessity of user-level long-term memory; however, the implementation details of the memory mechanism (storage structure, privacy boundaries, forgetting mechanism) are not publicly disclosed, marked [To be filled].

5.5. Evaluation & Observation Layer

The rehabilitation business's clinical evaluation system (outcome measurement of rehab training) is Fourier's ready-made L5 methodological asset, but the GR-3 humanoid product line itself has no publicly disclosed evaluation benchmarks, success rates, or reliability data. The CES / WAIC demos are capability showcases rather than quantitative evaluation.

5.6. L6 Governance & Safety Layer

Fourier's strongest L6 evidence is medical-grade compliance: the GRx series' clinical rehabilitation application at Perkeso (Malaysia's national social security organization) rehabilitation center — a robot entering a clinical environment must meet dual compliance for medical devices and institutional safety. Soft shell + pressure sensors + fast-system safety arbitration are engineering evidence of runtime safety. Governance details on data privacy (patient data, home-scenario imagery) are not disclosed.

5.7. Six-Layer Strengths & Weaknesses Summary

6. Real-World Cases

6.1. Malaysia PERKESO Rehabilitation Center

In 2026-06, Malaysia's PERKESO (national social security organization) rehabilitation center opened, bringing the GRx series into clinical rehabilitation applications, with GR-3 completing an eldercare cognition-training demo. This is the first humanoid-robot case among Chinese vendors in this group to enter the clinical environment of another country's national administrative institution; its significance lies in the replicability of clinical compliance: once institutional-level validation is passed in one country, the solution can be promoted to other medical institutions.

6.2. CES 2026 & WAIC 2026 Showcase

CES 2026 (2026-01-06~09): Fourier exhibited independently, showcasing multimodal interaction and a desktop-level concept prototype, marking its move from domestic trade shows to the global consumer-electronics stage. WAIC 2026 (2026-07): GR-3 demoed home-scenario tasks (guest reception, object delivery, home security, recognizing vague commands like "I'm thirsty"), and released the "Cat-head" robot family IP (GR Mini public-space interaction / GRNano desktop companionship) and the "Embodied Home" overall solution. Both show lines point to the same judgment: Fourier puts the humanoid robot's primary battlefield at the interface of human emotional connection and care, not factory cycle time.

7. Summary

7.1. Strengths & Limitations

Strengths: ten years of rehabilitation delivery experience and clinical-compliance assets; a Care-bot differentiation positioning that avoids the industrial red ocean; a dual-route cognition architecture that is a mature engineering solution for interaction scenarios; and a safety design of soft shell + pressure sensors that constitutes a hard-constraint advantage in "humans and robots sharing space" scenarios.

Limitations: opaque shipment and funding status (Omdia's ~2% share is an estimate); the humanoid product line's evidence of work capability (payload, precision, long-horizon tasks) is weaker than industrial vendors'; GR-3 pricing is undisclosed; and no named customer case has yet appeared for home-scenario commercialization.

7.2. Applicability Boundaries

Suitable for eldercare institutions, rehabilitation hospitals, high-end senior-care services, and public-space reception scenarios. Not suitable for factory cycle-time work (compared with UBTech / Galaxy General) or logistics sorting (compared with Figure). If a buyer's primary constraint is "safe cohabitation with people" rather than "work efficiency," Fourier is the best-matching candidate in this group.

7.3. Selection Recommendations

1. Eldercare institutions should prioritize evaluating the combined GR-3 + Intelligent Rehab Port solution and require institutional-integration experience from the PERKESO case.
2. Clarify the division of care-boundary responsibility before purchase: the Care-bot assists rather than replaces caregivers, and acceptance criteria should center on interaction safety and task completion rate.
3. Require the vendor to disclose memory and privacy design (face-data storage, home-imagery boundaries) — the L6 gap in medical and home scenarios is a compliance risk.
4. For industrial needs, the GR-3C variant can be evaluated, but its work-capability evidence is limited and should be tested head-to-head in the same venue with industrial vendors.

Information Gap Declaration

1. Shipment volume and funding status: official shipment figures not disclosed; Omdia's ~2% share is an estimate based on the ~13,000-unit global base in 2025 (a few hundred units), not an official disclosure; information on the latest public funding round is missing, all marked [To be filled].
2. GR-3 pricing not disclosed, marked [To be filled].
3. Historical GR-1 / GR-2 sales and rehabilitation revenue have no authoritative figures.
4. Memory and privacy mechanisms (L4 / L6 data side) implementation details not disclosed.
5. The specific number of overseas medical-institution deployments and the list of countries are only covered by the official statement of "multiple countries and regions," with no details.

8. References

1. Fourier official website (GRx / GRW / RehabHub product matrix) — Fourier Intelligence, 2026. https://www.fftai.cn
2. GR-3 series product page — Fourier Intelligence, 2026. https://fftai.com/grx/gr3
3. GRW wheeled dual-arm robot product page — Fourier Intelligence, 2026. https://fftai.com/grw
4. Intelligent Rehab Port solution — Fourier Intelligence, 2026. https://fftai.com/rehab/solution
5. GR-3 entry (WAIC 2026 / CES 2026 / PERKESO) — Baidu Baike (ima repost), 2026. <https://ima.qq.com/wiki/?shareId=5290662a81503203bcfb3020826556eca932dad8e8508405b17703b02baae921&mediaId=weburl_05c1dd48786ba2ac105d0edf0149c30f_7b96ce21ee3dcbcd5b7139ffbeabf35a7460976722994802>
6. GR-3 series entry — Baidu Baike, 2026. https://baike.baidu.com/item/GR-3%E7%B3%BB%E5%88%97/67514274
7. Fourier GR-3 (GRx Series) — The Robot Group, 2026. https://www.therobotgroup.co.uk/robot/fourier-gr3
8. Zhiyuan: full-stack leader in embodied intelligence (basis for Omdia's 2025 global humanoid shipment share) — Zheshang Securities, 2026-03-28. https://www.sgpjbg.com/info/df7be3e7242d311ece3785906662f44c.html
9. AI Robot Weekly — NetEase News, 2026-07. https://c.m.163.com/news/a/L1VPHI9505118I96.html