CtrlK
| Fourier GR-1 |
|---|
 |
| General information |
| Manufacturer |
| Country of origin |
| Release year |
| Status |
| Price |
| Availability |
| Website |
The Fourier GR-1 is a general-purpose humanoid robot developed by Fourier Intelligence (now branded as Fourier), a Shanghai-based robotics company with deep roots in rehabilitation robotics. Unveiled in July 2023 at the World Artificial Intelligence Conference (WAIC) in Shanghai, the GR-1 became one of the first humanoid robots in China to enter mass production, with over 100 units delivered to research institutions, enterprises, and pilot deployment sites by early 2024. Standing 165 cm tall and weighing 55 kg, the robot features more than 40 degrees of freedom, walks at speeds up to 5 km/h, and can carry payloads of up to 50 kg, nearly its own body weight.
The GR-1 occupies a distinctive position in the humanoid robotics landscape because of Fourier's background in medical rehabilitation technology. Unlike competitors that approach humanoid design from an industrial automation or consumer electronics perspective, Fourier drew on nearly a decade of experience building exoskeletons, robotic gloves, and other rehabilitation devices used in hundreds of hospitals worldwide. This heritage gives the GR-1 a particular emphasis on safe human interaction, biomechanical precision, and applications in healthcare and eldercare.
Fourier Intelligence was founded in 2015 by Alex Gu (Gu Jie) and Zen Koh in the Zhangjiang High-Tech Park in Shanghai's Pudong district. The company takes its name from Joseph Fourier, the French mathematician and physicist known for Fourier analysis. Alex Gu, born in 1981, is a graduate of Shanghai Jiao Tong University with a degree in mechanical engineering and previously worked as a sales executive at National Instruments. Zen Koh serves as co-founder and chief strategy officer.[1][2]
From its inception, Fourier concentrated on intelligent rehabilitation robotics. The company's early products included the ExoMotus M4, a lower-limb powered exoskeleton designed for patients recovering from stroke, spinal cord injury, and other central nervous system disorders. Fourier subsequently expanded its rehabilitation portfolio to include the ArmMotus (upper-limb rehabilitation), WristMotus (wrist therapy), AnkleMotus (ankle therapy), and the RehabHub platform, which integrates multiple therapy devices for comprehensive physical rehabilitation programs.[3]
By 2019, Fourier's rehabilitation robots had been deployed in approximately 500 hospitals and medical care centers across more than 20 countries, providing over 100 million interactive training sessions. That same year, the company began its pivot toward general-purpose humanoid robotics, recognizing that many of the core technologies developed for rehabilitation, including high-performance actuators, biomechanical modeling, and human-safe control systems, could be applied to a broader class of robots.[2][4]
In June 2020, Fourier acquired Zhuhai RHK Healthcare, a rehabilitation services provider serving over 200 hospitals across China, further solidifying its position in the healthcare robotics ecosystem.[5]
In July 2024, Fourier Intelligence formally rebranded, splitting into two distinct business units: Fourier, focused on general-purpose humanoid robotics (the GR series), and Fourier Rehab, a specialized subsidiary continuing the company's rehabilitation technology work. This restructuring allowed each unit to pursue focused development strategies for its respective market.[6]
Fourier initiated the humanoid robot program in 2019, working out of a small laboratory on the first floor of its Zhangjiang headquarters. The team leveraged years of experience in actuator design, sensor integration, and human biomechanics from the rehabilitation robotics division. As Alex Gu noted, "Many technologies used in rehabilitation robots are essentially applicable to humanoid robots."[2]
A significant milestone came in 2022 when Fourier's 1.65-meter prototype successfully stood upright and walked untethered for the first time. Gu later described the moment: "When we saw it standing up for the first time, untethered and walking around by itself, it was a big encouragement" for the engineering team.[7]
The GR-1 made its public debut at the 2023 World Artificial Intelligence Conference (WAIC) in Shanghai in July 2023, where it "instantly stole the show" according to media reports. The jet-black, lanky humanoid demonstrated bipedal locomotion, obstacle avoidance, slope navigation, and basic object manipulation. Some overseas viewers who saw promotional videos initially questioned whether the footage was computer-generated, prompting Fourier to release additional live demonstration footage.[7][8]
Fourier announced ambitious plans to manufacture 100 GR-1 units by the end of 2023, positioning the robot as one of the first mass-produced humanoid robots globally. The company clarified that these initial units would serve primarily as hardware and software development platforms distributed to R&D laboratories, universities, and AI companies rather than as finished commercial products.[9]
In December 2023, Chinese President Xi Jinping visited Shanghai for the first time since the pandemic and inspected Fourier Intelligence's facilities during a broader tour of the city's technology sector. According to reports, Xi asked whether it was possible to converse with the bipedal robot and have it perform basic tasks. The visit was widely interpreted as a signal of the central government's growing interest in humanoid robotics as a strategic technology.[10][11]
By early 2024, Fourier had delivered over 100 GR-1 units to companies and institutions across various sectors, providing the company with real-world feedback and insights for iterating on the design. These deployments spanned research universities, AI companies developing embodied AI systems, and enterprise pilot programs in banking and manufacturing.[12]
In June 2024, Fourier equipped the GR-1 with an upgraded vision system enabling real-time mapping, navigation, and obstacle avoidance capabilities, enhancing the robot's ability to operate autonomously in unstructured environments.[5]
The GR-1 is built around Fourier's proprietary FSA (Fourier Smart Actuator) technology and features a highly bionic, human-proportioned body designed for natural interaction with people and human-scale environments.
| Category | Specification | Value |
|---|---|---|
| Physical | Height | 165 cm (5 ft 5 in) |
| Physical | Weight | 55 kg (121 lb) |
| Mobility | Total degrees of freedom | 40-44 (varies by configuration) |
| Mobility | DOF per arm | 7 |
| Mobility | DOF per hand | 11 |
| Mobility | DOF per leg | 6 |
| Mobility | Max walking speed | 5 km/h (3.1 mph; 1.39 m/s) |
| Mobility | Max running speed | ~7.2 km/h (4.5 mph; 2 m/s) |
| Mobility | Stair climbing | Yes |
| Mobility | Slope navigation | Yes |
| Manipulation | Total payload capacity | 50 kg (110 lb) |
| Manipulation | Single-arm payload | 5 kg (11 lb) |
| Manipulation | Hand type | Dexterous, 5 fingers per hand |
| Actuators | Type | Fourier Smart Actuator (FSA) |
| Actuators | Peak joint torque | 230 N.m (general joints) |
| Actuators | Hip peak torque | 300 N.m |
| Computing | High-level computing | Intel Core i7-13700H |
| Sensors | Vision | 1x Intel RealSense depth camera + 6x RGB cameras |
| Sensors | Audio | Ring-shaped microphone array |
| Sensors | IMU | Fourier FLY108-IMU (proprietary) |
| Sensors | Force/torque sensors | Yes |
| Power | Battery | 11S1P 10Ah LiPo |
| Power | Estimated runtime | ~1-2 hours (task-dependent) |
| Connectivity | Interfaces | WiFi, Ethernet |
| Connectivity | API | Yes (proprietary SDK) |
| Connectivity | ROS compatible | Yes |
| AI | LLM integration | Yes (multimodal language model) |
| AI | Perception | BEV + Transformer + Occupancy Networks |
At the core of the GR-1's mechanical design is the Fourier Smart Actuator (FSA), a highly integrated joint module that consolidates the motor, driver, reducer, and encoder into a single compact unit. By designing and manufacturing its own actuators in-house, Fourier can optimize the cost-to-performance ratio of the entire system while reducing the physical size of the hardware compared to traditional servo-based approaches.[13][14]
The FSA modules are distributed across the GR-1's 40+ joints, with different variants tuned for the torque and speed demands of each joint location. The hip joint module produces the highest peak torque at 300 N.m, providing the power necessary for bipedal locomotion while carrying heavy loads. Other joints use FSA variants with peak torques up to 230 N.m, balancing between force output and the speed required for dynamic movements such as arm reaching, torso twisting, and head tracking.[9][14]
Fourier publishes documentation and control protocols for the FSA system, including the FSAControlWord communication protocol that allows developers to send calibration, motion control, and fault-handling commands to individual actuators. This openness supports the company's strategy of positioning the GR-1 as a research and development platform.[15]
The GR-1 employs what Fourier describes as a "pure vision perception solution" built on Bird's Eye View (BEV) representations, Transformer neural networks, and Occupancy Networks. Using input from six RGB cameras, the system generates a real-time 3D occupancy grid of the robot's surroundings, enabling autonomous navigation, obstacle avoidance, and spatial awareness.[14]
For human interaction, the GR-1 integrates a large language model (LLM) comparable in capability to ChatGPT, allowing it to understand and respond to natural language commands. Built-in emotional response systems and a high-resolution display enable social interaction, which is particularly important for the robot's healthcare and service applications.[14][16]
The robot's audio perception relies on a ring-shaped microphone array for voice recognition and sound source localization, while an Intel RealSense depth camera provides detailed 3D mapping of nearby objects and surfaces for manipulation tasks.[14]
Fourier trains the GR-1's locomotion and manipulation skills using NVIDIA Isaac Gym (and more recently, NVIDIA Isaac Lab), a GPU-accelerated physics simulation environment that allows thousands of parallel training instances to run simultaneously. This reinforcement learning approach enables the robot to develop robust motor skills in simulation before transferring them to real hardware.[17]
In one documented example, Fourier completed 3,000 training iterations for a floor-to-stand transition behavior in approximately 15 hours using Isaac Gym. When the resulting policy was transferred to the physical GR-1 hardware, it achieved an 89% success rate. The team also employs NVIDIA TensorRT for real-time inference optimization, CUDA libraries for parallel processing, and cuDNN for accelerating deep learning frameworks like PyTorch.[17]
The GR-1 targets a broad range of applications spanning healthcare, financial services, industrial manufacturing, and research.
Given Fourier's rehabilitation robotics heritage, healthcare represents a natural application domain for the GR-1. The robot has been tested assisting physical therapists by guiding patients through range-of-motion exercises and providing partial weight support during rehabilitation sessions. Its 50 kg payload capacity makes it suitable for patient transfer tasks, such as moving individuals between beds and wheelchairs. Fourier envisions the GR-1 serving as a companion robot for the elderly, particularly relevant given China's rapidly aging population and declining workforce.[9][16]
In one of the GR-1's most prominent commercial deployments, Fourier established the "GR-1 Bank Lobby Manager Scene Training Base" at the Shanghai Pudong Branch of China Construction Bank (CCB), one of China's largest state-owned commercial banks. At this deployment, the GR-1 serves as a lobby assistant, helping with business consulting, guiding customers to appropriate service counters, demonstrating how to use smart teller machines, and providing basic anti-fraud education. Fourier has indicated that future iterations will incorporate more integrated body language, walking-based customer greeting, lobby patrol capabilities, and potentially assistance with deposit, withdrawal, and account management tasks.[10][18]
By mid-2024, Fourier's humanoid robots were deployed at SAIC General Motors' (SAIC-GM) facilities, including the Jinqiao luxury vehicle assembly plant and the Ultium battery factory. At these sites, the robots perform tasks related to high-voltage component installation and precision assembly operations, demonstrating the GR-1 platform's potential in automotive production environments.[19]
A significant portion of GR-1 units have been distributed to universities, AI laboratories, and research institutions worldwide. The robot's compatibility with ROS (Robot Operating System), MuJoCo, and NVIDIA Isaac Lab makes it accessible to the broader robotics research community. Researchers use the platform for studies in locomotion control, dexterous manipulation, human-robot interaction (HRI), and embodied AI.[12][20]
The RoboCup Federation has partnered with Fourier (along with Booster Robotics and Unitree Robotics) to provide humanoid robot platforms at discounted rates for participating competition teams, further cementing the GR-1's role in academic robotics research.[20]
Fourier has identified additional target applications for the GR-1, including concierge and visitor guidance services, entertainment and exhibition demonstrations, industrial logistics, safety inspection, and household service and companionship.[12]
Fourier has not publicly disclosed an official retail price for the GR-1. However, third-party listings and industry estimates place the robot in the range of approximately $130,000 to $170,000 USD. One discontinued third-party listing priced the GR-1 at $129,950 USD.[21] Standard Bots and other industry sources have estimated the price at $150,000 to $170,000 for enterprise and research customers.[16]
The GR-1 is not available for general consumer purchase. Fourier distributes units primarily through direct partnerships with research institutions, universities, AI companies, and enterprise pilot programs. Interested parties must contact Fourier directly to discuss procurement.[12]
Fourier Intelligence has raised substantial venture capital across multiple funding rounds to support its transition from a rehabilitation robotics company to a general-purpose humanoid robotics leader.
| Round | Date | Amount | Lead Investor(s) |
|---|---|---|---|
| Series A | February 2018 | ~$4.8 million | Undisclosed |
| Series B / B+ | May 2020 | ~$14.1 million | Undisclosed |
| Series C | July-October 2021 | ~$14.9 million | Prosperity7 Ventures, Vision Plus Capital |
| Series D | January 2022 | RMB 400 million (~$63 million) | SoftBank Vision Fund 2, Prosperity7 Ventures |
| Series E | January 2025 | RMB | Guoxin Investment, Pudong Venture Capital, Prosperity7 Ventures |
The Series D round in January 2022, led by SoftBank Vision Fund 2 with participation from Prosperity7 Ventures (Saudi Aramco's venture arm) and Vision Plus Capital, set a record as the largest single funding round in the rehabilitation robotics sector at that time.[22][23]
The Series E round in January 2025, securing nearly $109 million, represented Fourier's largest funding round to date. Key investors included Guoxin Investment, Shanghai Pudong Venture Capital, and returning investor Prosperity7 Ventures.[24]
According to various sources, Fourier has raised between $193 million and $246 million in total funding across all rounds. The company's pre-money valuation reached approximately $1.1 billion (RMB 8 billion) by 2025, securing its status as a unicorn and one of the most highly valued humanoid robotics companies globally. Other notable investors across Fourier's funding history include IDG Capital and various Shanghai-based government investment vehicles.[25][26]
The GR-1 competes in an increasingly crowded humanoid robotics market, particularly within China, where government policy support and manufacturing infrastructure have fueled rapid development.
| Company | Robot | Key Differentiator |
|---|---|---|
| Fourier | GR-1 / GR-2 | Rehabilitation robotics heritage; FSA actuators; healthcare focus |
| Tesla | Optimus | Manufacturing scale; vertical integration; automotive AI transfer |
| Unitree Robotics | G1 / H1 | Aggressive pricing ($16,000-$90,000); high volume shipping |
| UBTECH | Walker S | Automotive factory deployments (BYD, Geely, Foxconn) |
| Agility Robotics | Digit | Warehouse logistics; Amazon partnership |
| Figure AI | Figure 02 | BMW manufacturing partnership; OpenAI integration |
Fourier's primary competitive advantage lies in its rehabilitation robotics expertise, which provides differentiated capabilities in healthcare applications and human-safe interaction. The company's in-house FSA actuator technology also provides cost and performance optimization advantages. However, competitors like Unitree have moved aggressively on pricing, with the Unitree G1 available at approximately $16,000, roughly one-tenth the estimated cost of the GR-1. In the automotive manufacturing segment, UBTECH's Walker S1 has secured larger deployment contracts with major Chinese automakers.[27]
Alex Gu has dismissed concerns about an emerging price war in humanoid robotics, stating that the industry remains "too immature for such competition." He has attributed the rapid momentum in Chinese humanoid robotics to two factors: advances in AI and the strength of China's automotive and consumer electronics manufacturing supply chains.[12]
On September 30, 2024, Fourier unveiled the GR-2, the second generation of its GRx humanoid robot series. The GR-2 represents comprehensive upgrades over the GR-1 across nearly every dimension.
Key improvements include a taller frame (175 cm, up from 165 cm), increased weight (63 kg), expanded degrees of freedom (53, up from 40), and upgraded FSA 2.0 actuators with peak torques exceeding 380 N.m. The GR-2's hands feature 12 degrees of freedom (compared to 11 in the GR-1) along with six array-type tactile sensors per hand that can sense force and identify object shapes and materials. The joint configuration was redesigned from a parallel to a serial structure, simplifying control, reducing manufacturing costs, and improving the transfer of skills learned in AI simulation to real-world operation.[28][29]
The GR-2 also features a detachable battery with approximately twice the capacity of the GR-1, providing up to two hours of continuous operation, and an improved software development kit supporting frameworks such as NVIDIA Isaac Lab and MuJoCo.[28]
Fourier subsequently expanded the GRx series further with the GR-3 (announced August 2025), which is explicitly designed for social care applications and features 55 degrees of freedom, a dual hot-swappable battery system with up to three hours of runtime, and an integrated emotional AI module. The GR-3 is priced at approximately $27,500 for B2B customers in China, representing a significant cost reduction from earlier models.[30]