The Fourier GR-2 is a full-size humanoid robot standing 175 cm tall and weighing 63 kg, built by Shanghai-based Fourier Robotics (formerly Fourier Intelligence). It features 53 degrees of freedom, 12-DOF dexterous hands with six array-type tactile sensors, and seven distinct FSA 2.0 actuator types delivering up to 380 N·m of peak joint torque — the highest in its class. Unveiled in September 2024 as the successor to the GR-1, the GR-2 targets healthcare, research, and industrial automation. It supports VR remote control, lead-through programming, and direct command modes. Pricing is estimated at $150,000+ for enterprise customers, with availability through Fourier's direct sales channels. Fourier has also launched the GR-3 "Care-Bot" (August 2025), a care-focused variant with soft-touch design and emotional interaction, which debuted at CES 2026.
Price and Availability
The GR-2 is available through enterprise engagement with Fourier Robotics. No retail pricing is published.
| Model | Est. Price | Target Market |
|---|---|---|
| GR-2 (enterprise/research) | $150,000+ (estimated) | Universities, research labs, healthcare, industrial |
| GR-3 Care-Bot (165 cm, 55 DOF) | Lower than GR-2 (not confirmed) | Elder care, education, companionship, public venues |
| GR-1 (predecessor, 165 cm, ~40 DOF) | Lower tier | Basic research, locomotion studies |
| Milestone | Date |
|---|---|
| GR-1 launched (first mass-produced Chinese humanoid) | Late 2023 |
| GR-2 unveiled | September 2024 |
| GR-3 "Care-Bot" unveiled | August 2025 |
| GR-3 debuted at CES 2026 | January 2026 |
| Fourier N1 open-source humanoid launched | Early 2025 |
| Series E funding (~$110M from Guoxin, Prosperity7/Saudi Aramco) | Early 2025 |
| Total funding raised | ~$193 million |
| Key investors | SoftBank Vision Fund 2, Prosperity7 Ventures (Saudi Aramco) |
| Rebrand from "Fourier Intelligence" to "Fourier" | Mid-2024 |
Full Specifications
GR-1 → GR-2 → GR-3 Evolution
| Spec | GR-1 (2023) | GR-2 (2024) | GR-3 (2025) |
|---|---|---|---|
| Height | 165 cm | 175 cm | 165 cm |
| Weight | ~55 kg | ~63 kg | ~71 kg |
| DOF | ~40 | 53 | ~55 |
| Hand DOF | Basic / none | 12 per hand | 12 per hand |
| Peak joint torque | Lower | 380 N·m | Not confirmed |
| Actuators | FSA 1.0 | FSA 2.0 (7 types) | Next-gen |
| Battery | Detachable | Swappable, ~2 hours | Hot-swappable, ~3 hours |
| Market focus | Research/locomotion | Research/industrial/healthcare | Care/companionship/emotional AI |
| Exterior | Industrial | Industrial | Soft-touch, approachable design |
GR-2 Chassis and Build
| Spec | Value |
|---|---|
| Height | 175 cm (5 ft 9 in) |
| Weight | ~63 kg (139 lbs) |
| Total DOF | 53 |
| Hand DOF | 12 per hand (24 total) — doubled from GR-1 |
| Arm payload | 3 kg per arm |
| Actuators | FSA 2.0 series — 7 distinct types, each tailored to specific joint torque demands |
| Peak joint torque | 380 N·m |
| Frame | Integrated cabling layout — compact, modular, serviceable |
GR-2 Hands
The GR-2's 12-DOF dexterous hands are the platform's standout feature. Each hand has five articulated fingers designed to mirror human physiology, equipped with six array-type tactile sensors that sense force, identify object shapes and materials, and adjust grip in real time. The hands support VR remote control, lead-through programming, and direct command — enabling comprehensive data capture from motion paths to tactile responses for bridging sim-to-real gaps.
Battery and Power
| Spec | Value |
|---|---|
| Battery | Swappable lithium |
| Runtime | ~2 hours |
| Swap design | Quick-change for minimal downtime |
Sensors
| Feature | Value |
|---|---|
| Tactile sensors (hands) | 6 array-type per hand — force sensing, shape/material identification |
| Vision | Cameras for perception and navigation |
| IMU | Yes |
| Additional | Joint encoders, force-torque across joints |
Computing and Software
| Feature | Value |
|---|---|
| SDK compatibility | NVIDIA Isaac Lab, ROS, MuJoCo |
| Control modes | VR remote control, lead-through programming, direct command |
| Data collection | Records motion paths, tactile responses, and operational data for AI training |
| AI approach | Imitation and reinforcement learning for human-like motion |
| Developer access | Open platform for research and AI development |
What Makes Fourier Different?
Rehabilitation Robotics Heritage
Before building humanoids, Fourier spent nearly a decade building rehabilitation exoskeletons and therapy robots deployed across 2,000+ medical institutions in 40+ countries. This clinical background gives Fourier unique expertise in human biomechanics, smooth motion control, and safety for human-proximate operation. Industry observers note that Fourier's humanoids move with "unusually smooth, human-like gait" — a direct result of their rehab robotics DNA.
Rapid Hardware Iteration
Fourier went from GR-1 to GR-2 to GR-3 in under two years — one of the fastest iteration cycles in the humanoid industry. Each generation serves a different purpose: GR-1 was the proof of concept, GR-2 is the high-torque research/industrial workhorse, and GR-3 is the care-focused consumer platform.
The GR-3 "Care-Bot" (August 2025)
The newest model shifts Fourier's focus toward emotional intelligence and care applications. At 165 cm and ~71 kg with ~55 DOF, the GR-3 features soft-touch materials, an animated facial interface, 31 pressure sensors for full-body touch detection, a 4-mic audio array, RGB + structured-light cameras, and a dual-path response system (fast reflexes for simple cues + contextual reasoning via large AI models). It debuted at CES 2026 and targets elder care, education, companionship, and public-facing service roles.
Buyer's Guide: What to Know Before Buying a Fourier GR-2
Enterprise-Only Availability
The GR-2 is sold through Fourier's direct enterprise and research channels. There is no consumer retail option. Contact Fourier for pricing, configuration, and delivery timelines.
Strong Healthcare Credentials
If your deployment involves healthcare, rehabilitation, or elder care applications, Fourier's decade of clinical robotics experience is a genuine differentiator that most competitors lack. The GR-2 (for physical tasks) and GR-3 (for care/companionship) cover complementary healthcare use cases.
No North American Sales Channel
As of April 2026, Fourier does not have a dedicated North American sales or support infrastructure. The company has a Singapore entity and global rehabilitation device presence, but humanoid sales are primarily through direct engagement. For North American buyers, Unitree's lineup offers more accessible purchasing channels.
GR-2 vs GR-3: Different Products for Different Needs
The GR-2 is Fourier's high-torque research platform (175 cm, 53 DOF, 380 N·m). The GR-3 is the care-focused companion (165 cm, ~55 DOF, soft exterior, emotional AI). They are not interchangeable — choose based on whether your application requires physical capability or social interaction.
Fourier GR-2 vs Similar Robots
- GR-2 vs Unitree H2: The H2 ($29,900) is dramatically cheaper with 31 DOF and 360 N·m leg torque. The GR-2 ($150,000+ est.) has more DOF (53 vs 31), higher peak torque (380 N·m), and 12-DOF dexterous hands with tactile sensors. The H2 is accessible to a broader market; the GR-2 targets enterprise buyers with healthcare-specific needs.
- GR-2 vs Unitree H1: The H1 ($90,000) has the world-record walking speed (3.3 m/s) but fewer DOF (19–27) and limited arm manipulation. The GR-2 offers much better hand dexterity (12 DOF per hand) and higher peak torque (380 N·m vs 360 N·m). The H1 is for locomotion research; the GR-2 is for manipulation and healthcare.
- GR-2 vs Apptronik Apollo: Apollo has 71 DOF (vs 53) and active factory deployments with Mercedes-Benz, GXO, and Jabil. The GR-2 has higher peak joint torque (380 N·m) and healthcare heritage. Apollo targets logistics/manufacturing; GR-2 targets research/healthcare.
- GR-2 vs Sanctuary AI Phoenix: Phoenix uses 20-DOF hydraulic hands with 5 mN tactile sensitivity — the most sensitive in the industry. The GR-2 has more total DOF (53 vs 20–30) and higher torque but less hand sensitivity. Phoenix uses Carbon AI for rapid task learning (under 24 hours); GR-2 uses VR/lead-through/direct control modes.
- GR-3 vs 1X NEO: Both target home/care environments. NEO ($20,000) ships to consumers in 2026 with 75 total DOF and teleoperation support. The GR-3 has clinical heritage and 31 pressure sensors for full-body touch detection. NEO is further along in consumer availability; GR-3 has stronger healthcare positioning.
