Meet Alex Gu, Founder of Fourier, company behind the GR-3 Care-bot

[Chris]

Interview with CEO and founder of Fourier, Gu Jie (Alex Gu) from a few years ago during the prototyping of the first GRx series iteration.

How to get to the universal humanoid robot? Fourier Founder: Robot + Multimodal Large Model​

Zhang Jing, chief reporter of Pengpai News
2023-07-12 15:46
Source: Pengpai News

· Four years ago, when humanoid robots were deployed, it was still a cold plate. Even now, Gu Jie still believes that the failure rate of developing humanoid robots far exceeds the success rate. In 10 years, the market prospects for humanoid robots are huge, but they have not yet reflected their commercial value, and it takes a lot of courage to invest capital and manpower in research and development.

· Wheeled robots do not involve the transfer of the center of gravity. As long as the motor rotates, they can move forward. The walking of bipedal robots is a process from balancing to breaking balance to obtaining balance. It is difficult to develop bipedal, and the balance ability, efficiency and multi-joint control of humanoid robots are still challenges.

He is 1.65 meters tall, weighs 55 kilograms, and has a body freedom of 40. When such a universal humanoid robot stood in front of him, Gu Jie, the founder of Shanghai Fourier Intelligent Technology Co., Ltd., sighed that human data was stored in the form of "ACGT" (nitrogen-containing base) strings, and the living life seemed to be more and more programms. But robots have more and more the feeling of living beings.

This year's World Artificial Intelligence Conference released a number of major innovation debuts, including Fourier Intelligence's universal humanoid robot GR-1. On July 12, Fourier Intelligence officially released the general humanoid robot strategy. In an interview with Pengpai Technology (www.thepaper.cn), Gu Jie said that in the future, general humanoid robot intelligence will be divided into motion intelligence and cognitive intelligence. GR-1 is iterating balance and athletic ability. When the stability reaches normal driving without falling and the walking speed reaches 10 km/h, it can iterate to the next step: adapt to more environments.

Cognitive intelligence depends on the development of large models. This robot enthusiast who has twice thrown himself into the robot entrepreneurship track believes that the current large model is mainly based on the large language model, and there is still no large robot model, which has not achieved complete multimodality. In the future, only by achieving a robot + multi-modal large model can we realize a real general-purpose humanoid robot.

Two start-ups on the robot track

Gu Jie, who has been fond of mechanical and "hands-on" since he was a child, has been passionate about robots. He joined the school's robot team during his undergraduate years at Shanghai Jiaotong University. He and his team transcended conventional thinking and designed a pitching robot, which won the best creative award in the first national robot innovation competition. After graduating in 2003, he worked as an engineer at the National Instruments Company in the United States, engaged in technology and marketing.

Until 2008, because of an accidental opportunity, Gu Jie came into contact with robots again and started his first business. Although he also thought about developing humanoid robots at that time, considering the technical maturity, he temporarily gave up this direction and turned to rehabilitation robots. After 2008, Gu Jie stayed in the hospital for a period of time one after another, observing how patients see a doctor, hoping to improve efficiency through robots. At that time, the newly built rehabilitation department in China did not have good equipment. They could only help patients do boring training by manually moving their arms and thighs.

At the beginning of the business, the rehabilitation robot developed by the team was a large robot with a controller weighing hundreds of kilograms, which made it impossible for patients to wear and walk. However, with "pure liking", Gu Jie undertook some factory system integration orders to "transfuse blood" for his rehabilitation robot dream.

Seven years later, the company was acquired. But Gu Jie didn't want to let go of the robot, and the wearable miniaturized rehabilitation robot became his next target. He wants to develop an exoskeleton robot to help people with physical disabilities regain motor functions, just like French mathematician Fourier laid the mathematical foundation for signal processing. In 2015, he named it after the mathematician Fourier and established "Fourier Intelligence" in Zhangjiang, Shanghai.

After having an entrepreneurial experience, Gu Jie has made a good strategic plan for the second entrepreneurial: technology and products should make breakthroughs and form a closed business loop to survive. Develop products around customer needs and hospital scenarios, not because engineers like or are cool. Hemiplegia patients are paralyzed in the upper and lower limbs on the same side. When hospitals purchase upper limb robots, they will definitely purchase lower limb robots. For this reason, Fourier Intelligence has developed upper limb robots and lower limb robots at the same time to break through common technology.

In terms of control technology, unlike the high-speed movement of industrial robots, rehabilitation robots require softness and movement adaptability. At the same time, the joint explosive power of the rehabilitation robot should be strong and can achieve a variety of functions. Due to the low difficulty of upper limb robot technology, Fourier intelligent upper limb robot became a product before lower limb robot. The average cycle of making a generation of robots is 2-3 years. Lower limb robots have been "back and forth for several generations". Finally, in 2017, Fourier Intelligence released the first commercial lower limb exoskeleton robot in China.

At present, Fourier Intelligent's rehabilitation robot product matrix is basically formed, including upper limbs, lower limbs, movement and balance, and physical factor therapy. The sales and service network has covered more than 40 countries and regions, and has been stationed in more than 2,000 hospitals and institutions. Now 120 patients only need two therapists and 32 robots to complete the training, which greatly improves efficiency with artificial + intelligence. Gu Jie said.

Source: https://m.thepaper.cn/newsDetail_forward_23821551

 

[Chris]

Universal humanoid robot = motor intelligence + cognitive intelligence

The research and development of exoskeleton robots has precipitated the underlying technologies of sensors, modular design, integrated actuators, bionic mechanical structures, etc., coupled with the research and development team and entrepreneurial experience. "With this ability, why not do it yourself?" Gu Jie didn't want to watch other teams develop humanoid robots first. After a period of rudimentary exploration, in 2019, Fourier Intelligence officially launched the general humanoid robot project.

Gu Jie defined the universal humanoid robot as a robot that can adapt to complex environments, complete human tasks and movements, and use human tools. If the humanoid robot moves slowly, it can only be used in performance scenes. Fast movement can go deep into the logistics scene. Small body and flexible movement mean that it can adapt to the environment and shuttle freely. If the future humanoid robot has cognitive ability, it may become a teacher, a driver, a cook or a worker, and the application scenario is close to infinite possibilities.

Gu Jie believes that in the future, universal humanoid robot intelligence will be divided into motor intelligence and cognitive intelligence. Among them, robots with motor intelligence can parkour in the future, which is what Fourier intelligence will do in the future, that is, to develop the "cerebrain" and body of robots. Robots with cognitive intelligence have common sense and can identify the environment. This is what the general artificial intelligence model will do in the future, that is, to develop the "brain" of robots. The development of the brain will definitely lay the foundation for the task disassembly and completion of general tasks of robots.

The general-purpose humanoid robot GR-1 released by Fourier Intelligence is 1.65 meters tall, weighs 55 kilograms, has a body freedom of 40 degrees, a walking speed of up to 5 kilometers per hour, and a load of 50 kilograms. It adopts electric drive technology, has a highly bionic trunk configuration, anthropomorphic motion control, and has fast walking, Agile obstacle avoidance, steady up and downhill, impact interference resistance and other sports functions. Structural innovation gives the robot a dexterous body, and the innovation of control ability makes it learn a lot of movements. This is just the beginning. Next year, it will learn to jump high, long jump and run. When it reaches a certain motor ability, there are many application scenarios. Gu Jie said that if combined with cognitive intelligence, he can work with people to complete the action. After Optimizing The Analysis And Processing Ability, When People Are Issuing Instructions, Robots Can Understand The Semantics, Disassemble, Respond And Complete Certain Tasks Independently, Which Is Also The Key Breakthrough Of Fourier Intelligent's Next-Generation Upgrade Products.

At present, GR-1 is iterating balance and athletic ability. When the stability reaches normal without falling, and the walking speed reaches 10 km/h, it can iterate to the next step: adapt to more environments, "see the sofa can conveniently sit down and stand up, and do movements on the uneven road in the wild."

GR-1 is the first-generation robot development platform of Fourier Intelligence. It cooperates to develop humanoid robot application scenarios for enterprises with professional scenarios and AI development capabilities. In the future, our robot will be an open platform, just like we can develop apps on mobile phones. We hope that downstream scene companies will cooperate with us to develop application scenarios for humanoid robots.

Source: https://m.thepaper.cn/newsDetail_forward_23821551

 

[Chris]

Stick to doing boring things

Is humanoid robot a concept or a good business opportunity?

"Frankly speaking, the fact that I wanted to make a humanoid robot at the beginning did not consider the business purpose too much." When the humanoid robot was deployed four years ago, it was a cold plate. Even now, Gu Jie still believes that the failure rate of developing humanoid robots far exceeds the success rate. In 10 years, the market prospects for humanoid robots are huge, but they have not yet reflected their commercial value, and it takes a lot of courage to invest capital and manpower in research and development. At present, Fourier Intelligence has spread 70% of its resources on general-purpose humanoid robots.

Humans free their hands by standing up and can use tools. Gu Jie said that it is difficult to develop a bipedal robot to stand up and walk. The balance ability, efficiency and multi-joint control of humanoid robots are still challenges. Wheeled robots do not involve the transfer of the center of gravity. As long as the motor rotates, they can move forward. The walking of bipedal robots is a process from balancing to breaking balance to achieving balance, which requires the control system of the "cererebelle". At present, Fourier Intelligence captures and analyzes human whole body movements through sensors to form an action library of gait trajectory to train robots.

For Gu Jie, the most touching thing about the two start-ups is to insist on doing boring things. 99% of the time is doing boring things. The product needs to be iterated little by little, the drawings need to be drawn one by one, the screws need to be screwed one by one, the customers need to visit one by one, be able to get through the boring things, and the mentality should be stable.

At the same time, enterprises do not evaluate job titles and do not seek academic value. Therefore, start-ups not only need innovative technologies, but also must find a business entry point. Start-ups must find a business closed-loop model in the first 2-3 years. Research and development should be market-oriented and iterate products around customer needs. If the product has no characteristics, both To B and To C are the red sea. The red sea of To C is more serious, because there is no opportunity to communicate with customers and cannot open the window of customers' inner demand.

Recalling that the humanoid robot stood in front of him, Gu Jie was still shocked. "When the humanoid robot's whole body is lit up, it can stand in front of people and move, which is completely different from seeing toys. Human data is stored in the form of 'ACGT' (nitrogen-containing bases) strings, and living beings seem to be more and more programmed. But robots have more and more feelings of living things.

Gu Jie judged that humanoid robots can be used in dangerous scenarios such as power grids and mine inspections within 2-3 years, and commercial scenarios can also be realized within 3-5 years, but home scenarios still take 5-10 years. He hopes to produce robots with robots in the future. Gu Jie believes that the development speed of domestic robots will be amazing in the future. "Because China has engineer dividends, strong innovation ability, the ability to mobilize the whole industry chain and the ability to mass production, a large number of application scenarios can be developed in the next step of iterative robots after rapid landing."

Source: https://m.thepaper.cn/newsDetail_forward_23821551

 

[Chris]

Exclusive Interview with Fourier Intelligence’s Gu Jie: Becoming the First Company to Mass-Produce General-Purpose Humanoid Robots

Humanoid robots are the last mile for general artificial intelligence.
Author: Jin Wang

From Waseda University’s WABOT in 1972 to Tesla’s Optimus debut in 2022, human scientists have spent half a century developing humanoid robots. Over this period, humanoid robotics has evolved from the technology validation phase, represented by Honda’s ASIMO and Boston Dynamics’ Atlas, to the commercial exploration phase, epitomized by Tesla’s Optimus. A new era of robotics is dawning.

To explore this shift, Zhiding Technology has launched a special report series titled *The New Era of Robotics*, aimed at uncovering the technological evolution and innovation stories of global robotics teams. This article is the first in-depth piece of the series.

In 2016, Fourier Intelligence’s core R&D team debated a critical question about the company’s product roadmap: “Should we develop humanoid robots?” At the time, Fourier, just one year old, was already making strides in rehabilitation robotics. However, for Gu Jie, who had been researching robots with the Shanghai Jiao Tong University robotics team since 2002, creating a humanoid robot was a lifelong dream.

As the ultimate form of robotics, humanoid robots are an irresistible allure for robotics idealists. No robotics team has not dreamed of building one. Gu Jie admitted, “I’ve been thinking about this for nearly 20 years.” However, both the broader robotics industry and Fourier’s technical capabilities in 2016 suggested that humanoid robots were still premature. Even if developed, they lacked commercial viability. Reason told Gu to wait.

Thus, Fourier’s team devoted itself entirely to medical rehabilitation robots, launching upper-limb rehabilitation robots and lower-limb exoskeleton robots in the following years, becoming one of China’s first teams to commercialize exoskeleton robots for medical rehabilitation.

Humanoid robots returned to the agenda in mid-2019, when Fourier officially launched its humanoid robot project internally. Coinciding with this, the global humanoid robotics field entered a new cycle of commercial innovation.

Source: https://www.techwalker.com/2023/0706/3150710.shtml

 

[Chris]

The Three Years Without Humanoid Robots

In 2015, Fourier Intelligence was officially established in Shanghai. Prior to this, Gu Jie had already been immersed in the robotics field for many years. During his previous entrepreneurial venture, he successfully launched large-scale rehabilitation robots for the medical sector. This time, Gu aimed to develop smaller, more compact robots.

At that time, an Israeli robotics team called Rewalk had, just a year earlier, received FDA approval in the U.S. for their exoskeleton robot designed for paraplegic patients with spinal cord injuries, making it the first FDA-approved exoskeleton robot in the U.S. Inspired by this, Gu wondered if a similar robot could be created in China to help people with similar mobility impairments stand again.

Another factor that solidified Gu’s resolve to develop medical exoskeleton robots was his visits and research at rehabilitation centers in major hospitals in Shanghai and Beijing. There, he saw hundreds of patients in a single hospital unable to walk like others, undergoing monotonous and arduous rehabilitation training using traditional methods guided by therapists. “The emotional impact and shock of this experience are indescribable unless you’ve spent time in a hospital’s rehabilitation department,” Gu told TechWalker.

Developing the first-generation lower-limb exoskeleton robot took Fourier Intelligence three years. During this period, the R&D team focused on mastering force feedback technology, a critical component for exoskeleton robots. In treating patients with limb disabilities, therapists often manually assist with exercises, such as bending or stretching a patient’s legs, adjusting the force based on the patient’s condition and recovery stage. Skilled therapists can intuitively gauge the appropriate force. For a robot to perform this task, it must assess the patient’s movement strength and apply the correct amount of force to complete the rehabilitation action—a process reliant on force feedback technology.

Leveraging its deep expertise in medical rehabilitation robotics, Fourier Intelligence developed the Fourier M2 upper-limb rehabilitation robot in just one year by the end of 2016. This robot, featuring assist, active, passive, and resistance training modes, was quickly adopted by top Chinese hospitals such as Huashan Hospital and Ruijin Hospital upon its release.

Source: https://www.techwalker.com/2023/0706/3150710.shtml

 

[Chris]

Fourier Intelligence’s Upper-Limb Rehabilitation Robot Fourier M2

This robot later became the first product in Fourier’s “Smart Rehabilitation Hub” solution to achieve a “grand slam” of medical device registration certifications in the U.S., China, Europe, and Australia, marking it as a landmark product for Chinese rehabilitation robots entering the global market.

Following the upper-limb rehabilitation robot, in early 2017, Fourier released its first lower-limb exoskeleton robot, the Fourier X1. From that moment, the Chinese market finally had its own medical lower-limb exoskeleton robot.

Beyond commercial success, what left a deeper impression on Gu Jie was seeing patients using the Fourier X1 stand for the first time. “We found that the first thing these patients did after putting on our exoskeleton was to take photos with their friends and family. Standing and taking pictures with loved ones became their happiest moment,” Gu recalled.

**Exclusive Interview with Fourier Intelligence’s Gu Jie: Becoming the First Company to Mass-Produce General-Purpose Humanoid Robots**

Paralympic champion Yao Fang stood and walked using the exoskeleton robot.

However, Gu also recognized that, at the time, Fourier’s lower-limb exoskeleton lacked self-balancing technology, requiring patients to use crutches for support. Achieving a breakthrough in self-balancing technology to address this limitation became the next major challenge for Fourier Intelligence.

Source: https://www.techwalker.com/2023/0706/3150710.shtml

 

[Chris]

An Unexpected Encounter with Humanoid Robots

In June 2019, Fourier Intelligence officially launched its humanoid robot project, though it remained a confidential initiative within the company, known to only a few. Rather than rushing to build the hardware, the team first mapped out the core technologies required for humanoid robots: developing advanced actuators, multidimensional force sensors, controllers, and control algorithms.

Among these, actuators—one of the three core components of robots—are particularly critical for humanoid robots. Fourier began developing its proprietary actuator, codenamed FSA, in 2018, during the development of its second-generation lower-limb exoskeleton. The first-generation X1 used off-the-shelf actuators, which were bulky, with motors, drive boards, and reducers taking up significant space and making the robot heavier. In contrast, the FSA actuator integrated these components, with focused advancements in structure, motor drive, reducer, and control software. By the release of the second-generation exoskeleton, this integrated actuator became a key highlight, marking the start of Fourier’s robots featuring proprietary integrated actuators.

However, for humanoid robots, a single actuator design was insufficient. Gu Jie told TechWalker, “Humanoid robots require a variety of actuator types and sizes. We later developed multiple integrated actuators specifically for humanoid robots.”

**Exclusive Interview with Fourier Intelligence’s Gu Jie: Becoming the First Company to Mass-Produce General-Purpose Humanoid Robots**

Diverse actuators are just one aspect of the technical complexity distinguishing exoskeletons from humanoid robots. Gu identified two fundamental differences:

1. **Hardware Structure**: Humanoid robots mimic human anatomy, requiring a head, torso, and limbs to perform a wide range of movements. Exoskeletons, as wearable devices, use sensors to provide torque compensation for human joints, aiding movement or rehabilitation.

2. **Motion Control**: Humanoid robots require complex autonomous control systems to perceive and respond to their environment, making decisions based on sensor data. Exoskeletons rely more on human interaction, using sensors to detect motion and applying algorithms for torque compensation.

Despite these differences, exoskeletons and humanoid robots share similarities, such as the need for self-balancing to navigate complex terrains safely. As Fourier tackled self-balancing technology for exoskeletons, its advancements in supply chain capabilities, integrated actuators, and multidimensional sensors brought it closer to humanoid robotics. “People may not realize that a lower-limb exoskeleton is essentially a bipedal humanoid robot. What we’ve been doing has always been evolving toward the prototype of a humanoid robot,” Gu reflected.

For a team of robotics enthusiasts, converging with the ultimate form of robotics—general-purpose humanoid robots—at the right moment was a rare opportunity and a stroke of luck for Fourier Intelligence.

Source: https://www.techwalker.com/2023/0706/3150710.shtml

 

[Chris]

“What’s the Pace per Kilometer of Your Humanoid Robot?”

On June 13, 2023, Fourier Intelligence officially announced through its channels that its general-purpose humanoid robot would be released in July, accompanied by a promotional video showcasing a humanoid robot walking briskly.

**Exclusive Interview with Fourier Intelligence’s Gu Jie: Becoming the First Company to Mass-Produce General-Purpose Humanoid Robots**

In the announcement video from June 13, Fourier’s humanoid robot is seen striding confidently. The test video, timestamped July 2022, demonstrated its motion capabilities. Gu Jie noted, “This was the result of experiments from a year ago. Many companies may not achieve this level of motion capability even in the next year or two.” The visible dexterity is considered a key advantage of Fourier’s humanoid robot.

According to Zhou Bin, president of Fourier’s General Robotics Division, “Some AI companies couldn’t wait for the final version of our humanoid robot and have already purchased prototypes during development to test their AI algorithms.”

Unlike lower-limb exoskeleton robots aimed at the medical market, Fourier’s humanoid robot focuses on maximizing motion capabilities to create a general-purpose platform. Gu told TechWalker that the target users are AI companies with commercial potential and research teams familiar with practical applications.

From deeming humanoid robot technology immature in 2016 to entering the field in 2019, Gu observed two significant market shifts. First, through technical breakthroughs, dexterous humanoid robots became achievable. Initially, the 2019 humanoid project was intended as a long-term technical reserve, not for public release. However, the team’s rapid progress exceeded expectations. Gu revealed, “The walking and obstacle-navigation capabilities of our upcoming humanoid robot are only at 20-30% of our expectations, but we have enough technical reserves to iterate its motion capabilities to 60% in the coming years.”

**Exclusive Interview with Fourier Intelligence’s Gu Jie: Becoming the First Company to Mass-Produce General-Purpose Humanoid Robots**

**Obstacle-Navigation Capabilities of Fourier’s Humanoid Robot**

Despite currently having the motion abilities of a child, Gu believes that as foundational technologies improve, the industry will shift from focusing on joints and degrees of freedom to evaluating humanoid robots by human athletic standards. In future pitches, investors might ask: “What’s the pace per kilometer of your humanoid robot? How far can it jump? How high can it reach?” Only then will humanoid robots become a universal platform.

**Slope-Climbing Capabilities of Fourier’s Humanoid Robot**

The second major shift is the rise of large-scale AI models, which have brought powerful capabilities to humanoid robots.

**The Last Mile for General Artificial Intelligence**

After the emergence of large-scale AI models, Fourier’s engineers discovered they could integrate these models into their humanoid robot platform. For example, they loaded ChatGPT onto an upper-body humanoid robot, enabling it to express happiness or perform an Arabic-style dance, with responses visualized in 3D software. Zhou Bin noted, “Our engineers were thrilled with these AI models. We had a blast experimenting internally.”

**Exclusive Interview with Fourier Intelligence’s Gu Jie: Becoming the First Company to Mass-Produce General-Purpose Humanoid Robots**

**Using ChatGPT to Make the Robot Perform a “Street Dance”**

However, they acknowledged that combining large-scale AI with humanoid robots is still in its early stages. Zhou, referencing Plato’s Allegory of the Cave, explained: “Before, AI hadn’t truly seen the world. Now, with a humanoid robot as its body, AI can finally step out and explore.” This vividly illustrates embodied intelligence.

Gu sees the integration of large-scale AI and humanoid robots as a perfect match: “AI companies create the soul and brain, but they need a body to perceive and interact with the physical world. Humanoid robots are the last mile for general AI.” A smart brain combined with a dexterous body forms the ultimate humanoid robot.

As a result, Fourier repositioned itself in 2023 to focus on general-purpose humanoid robot platforms, aiming to provide AI companies with robot bodies, much like car bodies for autonomous driving companies. On July 6, 2023, Fourier unveiled the GR-1, a 1.65-meter, 55-kilogram humanoid robot with 40 degrees of freedom, peak joint torque of 300 Nm, a walking speed of 5 km/h, and a 50-kilogram payload capacity.

**Exclusive Interview with Fourier Intelligence’s Gu Jie: Becoming the First Company to Mass-Produce General-Purpose Humanoid Robots**

What sets the GR-1 apart from other Chinese humanoid robots is its mass-production capability. Gu stated, “Fourier aims to be the first Chinese company to mass-produce humanoid robots. The GR-1 will officially enter mass production and sales by the end of this year.”

Source: https://www.techwalker.com/2023/0706/3150710.shtml

 

[Chris]

Alex Gu interview.

To view this content we will need your consent to set third party cookies.
For more detailed information, see our cookies page.

Accept third party cookies