CtrlK
The Fourier GR-2 is a next-generation general-purpose humanoid robot developed by Fourier Intelligence (now operating as Fourier), a Shanghai-based robotics company. Unveiled on September 30, 2024, the GR-2 is the successor to the GR-1 and represents a significant upgrade in physical capability, dexterity, and software integration. Standing 175 cm tall with 53 degrees of freedom and equipped with the company's proprietary FSA 2.0 actuator system, the GR-2 was designed for applications spanning healthcare, eldercare, industrial assistance, rehabilitation training, and academic research.[1][2]
The GR-2 is part of Fourier's GRx series of humanoid robots, which also includes the GR-1 (2023), the open-source Fourier N1 (April 2025), and the care-focused GR-3 (August 2025). Fourier differentiates itself among humanoid robotics companies through its long-standing background in rehabilitation robotics, a business that generates revenue from over 2,000 clinical institutions in more than 50 countries and helps fund the company's humanoid R&D.[3]
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 the end of 2023, 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]
The GR-2 was officially unveiled on September 30, 2024, with Fourier describing it as "a new milestone in the field of humanoid robotics with remarkable upgrades across hardware, design, and software."[1]
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]
| Specification | Value |
|---|---|
| Height | 175 cm (5 ft 9 in) |
| Weight | 63 kg (139 lb) |
| Total degrees of freedom | 53 |
| Hand degrees of freedom | 12 per hand |
| Maximum walking speed | 5 km/h (3.1 mph) |
| Single-arm payload capacity | 3 kg (6.6 lb) |
| Peak joint torque | 380 N.m |
| Actuator type | FSA 2.0 (7 variants) |
| Battery runtime | Up to 2 hours |
| Battery type | Detachable / swappable |
| Tactile sensors | 6 array-type per hand |
| Control methods | VR teleoperation, lead-through programming, direct command |
| Software compatibility | ROS, NVIDIA Isaac Lab, MuJoCo |
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. Industry analysts have noted that 380 N.m is enough to lift patients into wheelchairs, move furniture, and operate industrial tools.[9]
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]
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. While the numerical 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]
Some sources describe the GR-2's hands as "quadrupling the dexterity of earlier models," though this likely refers to the combined effect of increased DOF and the addition of tactile sensing rather than a purely numerical comparison of degrees of freedom.[8]
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]
The GR-2 features a detachable battery with double the capacity of the GR-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]
The following table summarizes the key differences between the GR-1 and GR-2.
| Feature | GR-1 (2023) | GR-2 (2024) |
|---|---|---|
| Height | 165 cm | 175 cm |
| Weight | 55 kg | 63 kg |
| Degrees of freedom | 44 | 53 |
| Hand DOF | 11 | 12 per hand |
| Tactile sensors | None | 6 array-type per hand |
| Peak torque | 230 N.m | 380 N.m |
| Actuator generation | FSA | FSA 2.0 (7 variants) |
| Encoder system | Single | Dual (2x accuracy) |
| Battery | Non-swappable | Detachable / swappable |
| Battery runtime | Not disclosed | Up to 2 hours |
| Joint architecture | Parallel | Serial |
| Payload capacity | Up to 50 kg (total carry) | 3 kg per arm |
| Walking speed | 5 km/h | 5 km/h |
| Exterior design | Skeletal / exposed frame | Enclosed plastic bodywork |
| Software platform | Basic SDK | Enhanced SDK with ROS, Isaac Lab, MuJoCo support |
The GR-2 delivers meaningfully more torque (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]
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.
The GR-2 supports three primary methods for control and task programming:
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.
Fourier developed the GR-2 in collaboration with NVIDIA, 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 is described as "an open-source modular framework for robot learning."[11]
The GR-2 is designed for deployment across several domains:
The GR-2 has been deployed to research institutions including ETH Zurich and Carnegie Mellon University for human-robot collaboration research. The robot has also been demonstrated performing tasks at an SAIC-GM automotive facility in China, as shown in videos published on Fourier's official website.[9]
Fourier continues to produce the GR-2 on a small scale alongside its newer models. CEO Gu Jie has stated that "large-scale production could still be years away," while predicting that the broader humanoid robot industry would see a "tenfold increase in output" in the near term.[5]
Fourier has not published an official retail price for the GR-2. Industry estimates place the cost at approximately $150,000 or more per unit, based on the GR-1's projected pricing of $150,000 to $170,000. The robot is sold exclusively through enterprise and institutional sales channels; no consumer purchase option exists. For comparison, the subsequent GR-3 care-focused model (launched in August 2025) was priced above 200,000 yuan (roughly $27,500), reflecting a different market positioning and capability scope.[5][9][12]
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 50 countries, 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]
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.
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, featuring 55 degrees of freedom, a multimodal emotional engine, and a hot-swappable battery system providing up to three hours of runtime.[3][12][13]
Fourier has raised approximately $193 million in total funding. The most recent round, a Series E worth nearly 800 million yuan (approximately $109 million), closed in January 2025 and was led by Prosperity7 Ventures, the venture capital arm of Saudi Aramco. Earlier investors include SoftBank Vision Fund 2 (which led the Series D in January 2022), IDG Capital, and Guoxin Investment. This capital supports the continued development and commercialization of the GR-2, GR-3, and N1 platforms.[3][14]
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), 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 starts at approximately $16,000), 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]