Citation and evidence

Fourier GR-2

16 min full readUpdated 20 references

This article's verification

Report a problem with this article

Edit article
More

Use this article

Raw MarkdownExplore connections

Improve this page

Revision historyDiscuss this article

Topics

Humanoid RobotsRobotics

Cite this article

The Fourier GR-2 is a general-purpose humanoid robot unveiled on September 30, 2024 by Fourier Intelligence (now operating as Fourier), a Shanghai-based robotics company. The GR-2 stands 175 cm tall, weighs 63 kg, provides 53 degrees of freedom, and runs on Fourier's proprietary FSA 2.0 actuator system with seven actuator variants delivering peak torques exceeding 380 N.m. Its 12-degree-of-freedom hands carry six array-type tactile sensors each, and a swappable battery gives up to two hours of runtime. The successor to the GR-1, it was designed for healthcare, eldercare, industrial assistance, rehabilitation training, and academic research.[1][2]

"GR-2 is a big step into the future of humanoid robotics," said Alex Gu, CEO of Fourier, on the day of the launch. "We're passionate about building the most intuitive embodied agent for AI."[1]

The GR-2 is part of Fourier's GRx series of humanoid robots, which also includes the GR-1 (2023) and the care-focused GR-3 series (GR-3 and GR-3C, from August 2025); Fourier also makes the open-source Fourier N1 (April 2025) and, from July 2026, the GRW wheeled dual-arm robot. Fourier differentiates itself among humanoid robotics companies through its long-standing background in rehabilitation robotics, a business that serves more than 2,000 organizations and hospitals in more than 40 countries and regions (as of 2024) and helps fund the company's humanoid R&D.[3]

What is the Fourier GR-2?

The Fourier GR-2 is a next-generation, full-size humanoid robot and the second flagship in Fourier's GRx line. Relative to the GR-1, it adds 9 degrees of freedom (53 versus 44), grows 10 cm taller and 8 kg heavier, replaces the first-generation Fourier Smart Actuator with the FSA 2.0 family, and introduces tactile-sensing dexterous hands plus a hot-swappable battery. Fourier positions it as a development platform for embodied AI rather than a consumer device, sold exclusively through enterprise and institutional channels.[1][2][7]

At the unveiling, Fourier stated that "GR-2 marks a new milestone in the field of humanoid robotics," citing upgrades across hardware, design, and software.[1]

"Say hello to GR-2!", official video from Fourier, published September 30, 2024. Watch on YouTube.

Background and Development

Fourier Intelligence was founded in 2015 in Shanghai by Alex Gu (Gu Jie) and Zen Koh. The company initially focused on rehabilitation robotics, building powered exoskeletons and therapy devices for clinical use. In July 2023, Fourier entered the humanoid robotics field with the launch of the GR-1 at the World Artificial Intelligence Conference (WAIC) in Shanghai. By 2024, more than 100 GR-1 units had been delivered to companies, universities, and AI research labs for development and testing purposes.[3][4]

Following the GR-1 launch, Fourier's CEO Gu Jie publicly acknowledged that the GR-1 remained an incomplete product, identifying six areas requiring optimization: motor intelligence, cognitive intelligence, bionic design, dexterity, commercial viability, and user experience. Customer feedback from the 100+ GR-1 deployments directly informed the GR-2's design priorities.[5]

In July 2024, Fourier Intelligence underwent a strategic rebranding, splitting operations into two distinct brands: Fourier for humanoid robotics and general-purpose robots, and Fourier Rehab for rehabilitation technology. This corporate restructuring reflected the company's dual strategy of pursuing both embodied AI and continued leadership in clinical rehabilitation robotics.[6]

When was the GR-2 released?

The GR-2 was officially unveiled on September 30, 2024. Fourier said that "with remarkable upgrades across hardware, design, and software, GR-2 marks a new milestone in the field of humanoid robotics."[1] In 2024, Fourier said it would keep producing the GR-2 on a small scale rather than in full mass production.[5]

Technical Specifications

Overview

The GR-2 stands 175 cm (5 ft 9 in) tall and weighs 63 kg (139 lb), making it 10 cm taller and 8 kg heavier than the GR-1. It provides 53 degrees of freedom, a substantial increase from the GR-1's 44. The robot achieves a maximum walking speed of 5 km/h (3.1 mph) and offers a single-arm payload capacity of 3 kg (6.6 lb). A detachable battery with double the capacity of the GR-1 enables up to two hours of continuous operation.[1][2][7]

SpecificationValue
Height175 cm (5 ft 9 in)
Weight63 kg (139 lb)
Total degrees of freedom53
Hand degrees of freedom12 per hand
Maximum walking speed5 km/h (3.1 mph)
Single-arm payload capacity3 kg (6.6 lb)
Peak joint torque380 N.m
Actuator typeFSA 2.0 (7 variants)
Battery runtimeUp to 2 hours
Battery typeDetachable / swappable
Tactile sensors6 array-type per hand
Control methodsVR teleoperation, lead-through programming, direct command
Software compatibilityROS, NVIDIA Isaac Lab, MuJoCo

Expanded article table

FSA 2.0 Actuators

The Fourier Smart Actuator (FSA) is the company's proprietary actuator technology, integrating the motor, driver, reducer, and encoder into a single compact module. The GR-2 uses the second-generation FSA 2.0, which includes seven distinct actuator variants, each tailored to meet the specific torque demands of different joints throughout the robot's body. The FSA 2.0 system delivers peak torques exceeding 380 N.m, a significant improvement over the GR-1's 230 N.m hip actuators (some sources cite 300 N.m for the GR-1's peak).[1][2][8]

A key innovation in the FSA 2.0 is a dual-encoder system that doubles the control accuracy compared to the previous generation. This improvement enables more precise movements, which is particularly important for tasks requiring fine motor control in high-precision environments such as healthcare, rehabilitation assistance, and manufacturing.[1][7]

The torque output of the FSA 2.0 is sufficient for demanding physical tasks. The robot buyer's guide BotInfo says the 380 N.m torque suits furniture manipulation, parts handling, and tool operation, and lists patient transfers among the robot's healthcare uses.[9]

Joint Architecture

Fourier redesigned the GR-2's joint configuration, shifting from a parallel structure (used in the GR-1) to a serial structure. This architectural change simplifies the control system, reduces maintenance complexity, and lowers manufacturing costs. The serial joint design also improves the robot's ability to transition from AI simulation training to real-world applications, since serial structures are easier to model accurately in simulation environments.[1][2]

The GR-2 also features an integrated cabling design for power and communication transmission. Wires are concealed within the robot's frame, resulting in a more compact package that is easier to modularize and customize for specific applications. The exterior is encased in sleek plastic bodywork, a notable aesthetic improvement over the GR-1's more skeletal, exposed appearance.[7][8]

Dexterous Hands

One of the most significant upgrades in the GR-2 is its hand system. Each hand provides 12 degrees of freedom, compared to the GR-1's 11-DOF hands, which Fourier describes as "double the dexterity of previous models."[2] While the numerical DOF increase is modest, the more important advancement is the addition of six array-type tactile sensors per hand. These sensors enable the GR-2 to detect applied force, identify object shapes and materials, and adjust its grip in real time for optimal manipulation in dynamic settings.[1][2][10]

The hands are electrically actuated and designed to mirror the flexibility of human physiology. The combination of increased dexterity and tactile sensing allows the GR-2 to perform tasks that require nuanced manipulation, such as handling delicate objects, operating tools, and interacting safely with people in care environments.[7][10]

Interesting Engineering described the hands as "quadrupling the dexterity of earlier models," which does not match Fourier's own "double the dexterity" wording.[20]

Perception System

The GR-2 is equipped with cameras and depth sensors for environmental perception. While Fourier has not published a complete breakdown of the GR-2's sensor suite, the robot's SDK provides pre-configured modules for machine vision, path planning, and force feedback control. The GR-1 used a combination of Intel RealSense depth cameras and six RGB cameras with a BEV (Bird's Eye View), Transformer, and Occupancy Network architecture for 3D environmental mapping; the GR-2 is expected to use a similar or improved perception stack.[3][11]

Battery and Power

The GR-2 features a detachable battery with double the capacity of the GR-1, extending runtime by roughly one hour over the prior generation.[1] The swappable design allows operators to quickly replace a depleted battery and return the robot to operation without extended downtime. This practical improvement extends the robot's effective operational window beyond the two-hour single-charge runtime, an important consideration for commercial deployment scenarios that require sustained availability.[1][2]

How does the GR-2 differ from the GR-1?

The following table summarizes the key differences between the GR-1 and GR-2.

FeatureGR-1 (2023)GR-2 (2024)
Height165 cm175 cm
Weight55 kg63 kg
Degrees of freedom4453
Hand DOF1112 per hand
Tactile sensorsNone6 array-type per hand
Peak torque230 N.m380 N.m
Actuator generationFSAFSA 2.0 (7 variants)
Encoder systemSingleDual (2x accuracy)
BatteryNon-swappableDetachable / swappable
Battery runtimeNot disclosedUp to 2 hours
Joint architectureParallelSerial
Payload capacityUp to 50 kg (total carry)3 kg per arm
Walking speed5 km/h5 km/h
Exterior designSkeletal / exposed frameEnclosed plastic bodywork
Software platformBasic SDKEnhanced SDK with ROS, Isaac Lab, MuJoCo support

Expanded article table

The GR-2 delivers meaningfully more torque (a 65% increase over the GR-1's 230 N.m rating), improved dexterity through tactile sensing, a redesigned joint architecture for easier manufacturing and sim-to-real transfer, and a substantially longer battery life with hot-swap capability.[1][2][7][8]

Software and Development Platform

SDK and Framework Compatibility

Fourier optimized the GR-2's development platform by introducing a new software development kit (SDK) compatible with mainstream robotics frameworks. The SDK supports ROS (Robot Operating System), NVIDIA Isaac Lab, and MuJoCo, giving developers access to pre-optimized modules for machine vision, path planning, and force feedback control through intuitive APIs.[1][2]

The SDK is designed to lower the barrier to entry for researchers and commercial developers who want to build applications on the GR-2 platform. By supporting widely used open-source frameworks, Fourier enables integration with the broader robotics and AI development ecosystem.

Control and Programming Methods

The GR-2 supports three primary methods for control and task programming:

  • VR teleoperation: Operators can control the robot's arms and hands remotely using virtual reality equipment, enabling intuitive manipulation for complex or hazardous tasks.
  • Lead-through programming: An operator physically guides the robot's limbs through a desired motion path, and the system records the movement data. This is a common technique in industrial robotics for quickly teaching repetitive tasks.
  • Direct command: The robot can be controlled through standard programmatic interfaces and APIs.[1][7]

These teaching modes allow the GR-2 to record a comprehensive set of operational data, from motion paths to tactile responses. This recorded data can then be used for training machine learning models, enabling the robot to learn from human demonstrations.

NVIDIA Isaac Gym Integration

Fourier developed and tested the GR-2 using NVIDIA Isaac Gym for simulation-based reinforcement learning. By simulating real-world conditions in parallel virtual environments, Fourier's team was able to reduce physical testing time and cost substantially.[11]

During development, the Fourier team completed 3,000 training iterations in approximately 15 hours for a floor-to-stand maneuver. The trained model achieved an 89% success rate when its action tensors were transferred directly to the physical GR-2 robot, demonstrating effective sim-to-real transfer. The team leveraged NVIDIA TensorRT for real-time inference optimization, CUDA libraries for parallel processing, and cuDNN for accelerating deep learning frameworks like PyTorch.[11]

Fourier is currently porting its workflows from the deprecated NVIDIA Isaac Gym to NVIDIA Isaac Lab, which it describes as "an open-source modular framework for robot learning."[11]

What is the Fourier GR-2 used for?

Target Markets

The GR-2 is designed for deployment across several domains:

  • Healthcare and eldercare: Patient assistance, rehabilitation support, and companionship in hospitals and nursing facilities.
  • Academic and industrial research: Serving as a development platform for AI researchers and robotics engineers at universities and corporate labs.
  • Industrial assistance: Manufacturing environments where the robot's strength and dexterity can assist with assembly, inspection, and material handling.
  • Guided tours and service: Visitor guidance in commercial settings such as museums, exhibition centers, and corporate campuses.[1][2]

Known Deployments

According to BotInfo, GR-2 units have gone to research institutions including ETH Zurich and Carnegie Mellon University for locomotion, manipulation, human-robot interaction, and embodied AI research, and the robot has been demonstrated on manufacturing tasks at an SAIC-GM automotive facility in China.[9]

In 2024, Gu Jie told KrASIA that large-scale production could still be years away and that Fourier would keep producing the GR-2 on a small scale; he predicted that the humanoid robot industry would see a tenfold increase in output by 2025.[5]

How much does the Fourier GR-2 cost?

Fourier has not published an official retail price for the GR-2. At the GR-3 launch in August 2025, robotics writer Mike Kalil reported that Fourier's previous GR robots sell for more than $100,000.[19] The robot is sold exclusively through enterprise and institutional sales channels; no consumer purchase option exists. Fourier did not announce a price for the subsequent care-focused GR-3 (launched in August 2025) either; at launch, TMTPost estimated that its cost could be well above 200,000 yuan (about $28,000).[18]

Fourier's Broader Strategy

Rehabilitation Robotics Heritage

Fourier's approach to humanoid robotics is informed by its decade-long history in rehabilitation robotics. Since 2015, the company has developed a comprehensive portfolio of rehabilitation devices, including the ExoMotus M4 lower-limb exoskeleton, the ArmMotus EMU upper-limb rehabilitation robot, and specialized ankle and wrist therapy systems. These products are deployed in over 2,000 institutions across more than 40 countries and regions (as of 2024), generating revenue that helps fund humanoid robotics R&D.[3][6]

This rehabilitation heritage gives Fourier deep expertise in human biomechanics, safe human-robot interaction, and clinical-grade force control. These capabilities transfer directly to humanoid robot design, particularly for healthcare and eldercare applications where safe physical contact with people is essential.[3]

Dual-Brand Structure

The 2024 rebranding into Fourier (humanoid robotics) and Fourier Rehab (rehabilitation technology) formalized the company's dual-track approach. Fourier Rehab operates as a specialized subsidiary under the Fourier Group, with its own dedicated team. Zen Koh, the co-founder, serves as CEO of the global hub based in Singapore and oversees international operations for the rehabilitation business.[3][6]

This structure allows each division to focus on its core market while sharing fundamental technology, such as the FSA actuator platform, across both product lines.

GRx Series Roadmap

The GR-2 sits in the middle of Fourier's evolving humanoid product line. The company followed the GR-2 with the Fourier N1, an open-source humanoid robot released in April 2025 that stands 1.3 meters tall and has all hardware designs, CAD files, and software published on GitHub. In August 2025, Fourier launched the GR-3, its first "care-centric" humanoid specifically designed for social care and companionship, standing 165 cm tall with 55 degrees of freedom, a multimodal emotional engine, and a hot-swappable battery system providing up to three hours of runtime.[3][12][13]

Funding

On 7 January 2025 Fourier announced an E-series financing of nearly 800 million yuan (nearly $109 million), with investors including Guoxin Investment, Pudong Venture Capital, Prosperity7 Ventures (the venture arm of Saudi Aramco) and Junshan Capital.[14][15] Earlier, SoftBank Vision Fund 2 led a 400 million yuan Series D in January 2022.[17] In 2025 Zhejiang Runyang New Material Technology invested 300 million yuan at a pre-money valuation of 8 billion yuan, which implies a valuation of about 8.6 billion yuan (about $1.2 billion) at the round price.[16] The company's full funding history is covered in the Fourier Intelligence article.

Industry Context and Competition

Who competes with the Fourier GR-2?

The GR-2 entered a rapidly growing and increasingly competitive humanoid robotics market. In China, Fourier competes with Unitree Robotics (G1, H1, H2), UBTECH (Walker S series), and newer entrants such as Agibot. Internationally, it faces competition from Tesla (Optimus), Figure AI (Figure 02 and Figure 03), Boston Dynamics (Atlas), Apptronik (Apollo), and Sanctuary AI (Phoenix).[3]

Fourier's competitive differentiation centers on its rehabilitation robotics background, which provides clinical-grade expertise in safe human-robot interaction. The company's model of using rehabilitation revenue to fund humanoid R&D also offers a more sustainable funding approach than competitors that rely solely on venture capital.[3]

Among Chinese competitors, Unitree has gained attention for affordability (the G1 launched at approximately $16,000 and now lists from about $13,500), while UBTECH focuses on large-scale industrial deployments at facilities like Foxconn. Fourier's niche centers on healthcare, eldercare, and research applications where its rehabilitation heritage provides a natural advantage.[3]

Continue reading

More to read

Topics:Humanoid RobotsRobotics

References

  1. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11 ^12 ^13 ^14 ^15 ^16"Fourier Unveils the Next-Generation Humanoid Robot GR-2." PR Newswire, September 30, 2024. prnewswire.com/...on-humanoid-robot-gr-2-302262068
  2. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10 ^11"Fourier GR-2." Fourier (fftai.com), official product page (archived January 2025). web.archive.org/...products-gr2
  3. ^1 ^2 ^3 ^4 ^5 ^6 ^7 ^8 ^9 ^10"Fourier Intelligence." AI Wiki. aiwiki.ai/...fourier_intelligence
  4. ^"Fourier Intelligence launches production version of GR-1 humanoid robot." The Robot Report, December 2023. therobotreport.com/...rsion-of-gr-1-humanoid-robot
  5. ^1 ^2 ^3"Fourier Intelligence's founder breaks down what's next for humanoid robots." KrASIA, 2024. kr-asia.com/...down-whats-next-for-humanoid-robots
  6. ^1 ^2 ^3"Fourier Intelligence rebrands to Fourier and Fourier Rehab." The Robot Report, July 16, 2024. therobotreport.com/...brands-fourier-fourier-rehab
  7. ^1 ^2 ^3 ^4 ^5 ^6 ^7"Bigger, stronger, smarter GR-2 humanoid launches with major hand upgrades." New Atlas, September 30, 2024. newatlas.com/...fourier-gr-2-humanoid-robot
  8. ^1 ^2 ^3"Fourier launches GR-2 humanoid, software platform." The Robot Report, September 30, 2024. therobotreport.com/...2-humanoid-software-platform
  9. ^1 ^2"Fourier Humanoid Robots: GR-2, GR-3 Specs and Pricing." BotInfo, 2026. botinfo.ai/...fourier-humanoid-robots
  10. ^1 ^2"Fourier GR-2 Humanoid Robot Unveiled, Gets Upgraded 12-DoF Dexterous Hands." TechEBlog, October 2024. techeblog.com/...urier-gr-2-humanoid-robot-upgrade
  11. ^1 ^2 ^3 ^4"Spotlight: Fourier Trains Humanoid Robots for Real-World Roles Using NVIDIA Isaac Gym." NVIDIA Technical Blog, 2024. developer.nvidia.com/...les-using-nvidia-isaac-gym
  12. ^"Fourier Intelligence unveils the Care-bot GR-3." TechNode, August 7, 2025. technode.com/...ull-sized-companion-humanoid-robot
  13. ^"Fourier Launches First Open-Source Humanoid Robot, Fourier N1." AIBase, April 2025.
  14. ^"Chinese robotics firm Fourier books nearly $109m in Prosperity7-backed Series E round." DealStreet Asia, January 2025. dealstreetasia.com/...backed-series-e-round-425571
  15. ^"傅利叶完成新一轮融资,E系列融资近8亿元" (Fourier completes new financing; E-series nearly RMB 800 million). PEdaily (投资界), 7 January 2025. news.pedaily.cn/...545167.shtml
  16. ^Zhejiang Runyang New Material Technology Co., Ltd. (300920). "关于对外投资暨增资上海傅利叶智能科技有限公司的进展公告" (Progress announcement on the capital increase in Shanghai Fourier Intelligence Technology Co., Ltd.), 27 June 2025. file.finance.qq.com/...1223998765.PDF
  17. ^"孙正义最新一笔投资:傅利叶智能完成4亿D轮融资" (Fourier Intelligence completes RMB 400 million Series D). PEdaily (投资界), 26 January 2022. news.pedaily.cn/...485847.shtml
  18. ^"傅利叶发布陪伴式人形机器人GR-3" (Fourier releases companion humanoid robot GR-3). TMTPost (钛媒体), August 2025. tmtpost.com/7650333
  19. ^Mike Kalil, "Fourier's New GR-3 AI Care Bot Humanoid Robot." mikekalil.com, August 8, 2025. mikekalil.com/...fourier-gr3-carebot
  20. ^Jijo Malayil, "53 degrees of freedom: New humanoid robot offers more flexibility, 2x battery life." Interesting Engineering, October 1, 2024. interestingengineering.com/...raded-humanoid-robot

Improve this article

Add missing citations, update stale details, or write a clearer explanation. Anyone can edit: changes go live when you publish, and every edit is checked by an editor.

9 revisions · v9 · 3,146 words · full history

Fact-checks are independent of edits: a reviewer re-verifies the article against its sources and stamps the date. How we verify

Research and drafting on this wiki are AI-assisted, under named human editorial standards. How AI is used here

Reviewer note: Independent fact-check of the full article against its cited sources (85 claims, 24 sources fetched); 1 material + 19 minor fixed

Cite this page: AI Wiki. "Fourier GR-2." aiwiki.ai, updated 1 Nov 2026, fact-checked 2 Nov 2026. CC BY 4.0. https://aiwiki.ai/wiki/fourier_intelligence_gr_2

Edit article