Inside Fourier Intelligence’s giant leap from rehab robots into humanoids

Gu's vision for GR-1 is an Android-like open-source platform. Ecosystem partners will be able to advance innovation throughout the industrial chain and create synergies conducive to rising adoption of the device.

Fourier Intelligence (傅利叶智能), a robot company that recently shot to fame over its release of a general-purpose humanoid robot, announced its plan to venture into the trillion-dollar general robotics segment at a conference yesterday in Shanghai.

The Shanghai-based startup released a video of GR-1, its first humanoid, to much fanfare at the opening ceremoney of World Artificial Intelligence Conference (WAIC) 2023 on July 6.

The firm has been in the media spotlight lately, with lots of hype around its cutting-edge technologies, as well as curiosity about how it got here and where it is headed next.

Alex Gu, the founder and CEO of Fourier Intelligence, set up shop in 2015 and dreamed about building humanoids one day.

All photos courtesy of Fourier Intelligence

“As a college student, I looked forward to the day of constructing a general-purpose humanoid robot with versatile applications,” Gu explained. “Although the failure rate is high, the value [of doing this] would be immense.”

In the face of limitations such as the robot’s motion capacity, AI algorithms and low market adoption, he decided that the startup ought to start from a low base and set its sights on something more pragmatic and practical — rehabilitation robotics.

A circuitous approach

Fourier Intelligence unveiled the country’s first lower extremity exoskeleton in 2017 and commercialized it after multiple iterations.

The exoskeleton, which is electricity-powered and programmed to walk bipedally, became the foundation on which Gu and his co-workers base their humanoid ambitions.

Conditions didn’t become ripe until in 2019, when the humanoid project code-named GR-1 launched internally — and quietly.

Everything about GR-1 was kept under wraps until it made its debut in early July.

All photos courtesy of Fourier Intelligence

Gu attributed the decision to keep a low profile to the pervasive skepticism about humanoids then.

“Nobody was upbeat about the prospects of a bipedal humanoid and building one seemed like a fool’s errand,” he told the audience yesterday. “We didn’t want our engineers to be distracted or discouraged by naysayers.”

The hush-hush over GR-1 paid off. Lab engineers were given a free rein to explore their options, devoid of unwanted attention or influences from outside the company.

When a prototype did come out and started walking, however clumsily in the beginning, Gu said his engineers were overwhelmed by euphoria, as if they created a new species.

Past experience from building mass-produced exoskeletons proved vital to the birth of a humanoid, thanks to the shared modular design, whereby parts can be joined and removed flexibly.

“This is like building Lego bricks, one piece at a time,” said Gu. “Very early on we began to assemble core components of a humanoid.”

Brawn and brains

After almost six years of development, Fourier Intelligence put together its proprietary technologies in sensor, all-in-one actuator, modular design, mechanical bionic structure, and rolled out a robot with a human-like figure.

A consensus across the world is that humanoids with multi-purpose functions need strong motion control as well as powerful cognitive capabilities.

In the terminology of robotics, a humanoid needs to have athletic intelligence and cognitive intelligence to be taken seriously.

All photos courtesy of Fourier Intelligence

These two attributes took time — and efforts — to develop. Six years ago, when Fourier Intelligence put its wearable exoskeleton on a mobility-impaired user, the kinetics of each joint couldn’t generate enough power to support the weight of the device and of the user per se.

Instead, the user had to rely on crutches for balance.

Now GR-1 has 40 degrees of freedom, coupled with a peak torque of 300NM produced by a joint module at the hip.

Better still, the self-balancing robot can carry a load of 50 kg and walk, avoid obstacles, navigate slopes and withstand disturbances — all on its own.

In video footage shown by Gu at the conference, the robot is even seen righting itself after slipping on a slope.

Although it might still be a long way before GR-1 can replicate acrobatic moves like those stunts by Boston Dynamics’ Atlas, the inroads Fourier Intelligence engineers have made seem to offer a cause for celebration.

But Gu decided it was too early to be smug about the achievements.

Logical progression

He’s now eyeing the next rung up the ladder for the gadget, in particular a “gigantic” breakthrough in both athletic and cognitive intelligences.

“Hopefully, our next-gen humanoid can run, leap forward and jump up high,” he said.

In the meantime, efforts are needed to optimize the analytical capacity of the robot, preferably through introduction of a large language model.

All photos courtesy of Fourier Intelligence

Therefore, once a command is issued, it will be able to understand the semantics, break down and respond to a task and even solve it independently, Gu noted.

Despite widespread pessimism toward humanoid robots, Saudi Aramco’s Prosperity7 Ventures (P7VC for short), a Series C-stage investor in Fourier Intelligence, has continued to support its bid to develop one.

During an exclusive interview with cnrobopedia.com yesterday, Aysar Tayeb, executive managing director of P7VC, said the venture fund doesn’t see this humanoid endeavor as a “branching out,” but as “a normal logical progression.”

He said Fourier’s foray into humanoid robotics aligns with the strategy of his fund, which is “investing always for the purpose of finding the most disruptive technologies that will have a positive impact on people.”

Changing mindsets

A race is on worldwide to produce humanoid robots, with stakes getting higher and higher after the entry of giants like Tesla, OpenAI and others.

The exponential growth of AI, typically generative AI, is believed to be a game-changer in the fortunes of humanoid robots, once an object of ridicule in the eyes of Dr. Kai-Fu Lee for a dearth of application scenarios.

Tides have turned. AI needs a carrier to better interact and evolve with the physical environment.

Against this backdrop, humanoid and generative AI is a “match made in heaven” and will address the “last-mile” issue in realizing artificial general intelligence (AGI), or embodied AI, Gu said.

All photos courtesy of Fourier Intelligence

The entrepreneur already has a plan. The focus will be on building an ecosystem surrounding Fourier Intelligence’s humanoid.

The first step toward constructing such an ecosystem, according to him, is to open an API for integrators or developers with specific industrial know-how to find an array of use cases, like security inspection, elderly care and household services.

Gu’s vision for GR-1 is an Android-like open-source platform. Ecosystem partners will be able to advance innovation throughout the industrial chain and create synergies conducive to rising adoption of the device.

“We as a company cannot do it alone; rather, we need partners to join us on this journey,” said Gu. “Humanoids are not a consumable that will become mass produced and utilized within one year.”

But one is equally wrong to write it off as some sci-fi fantasy that will come true only in five to ten years, he cautioned.

Avatar photo
Ni Tao

Ni Tao is the founder and editor-in-chief of cnrobopedia. Prior to cnrobopedia, he had a full decade of experience with a major state-run English-language newspaper as a tech reporter and opinion writer. He is also a communications specialist, having provided consultancy services to established firms like Siemens, Philips, ABinBev, Diageo, Trip.com Group (Nasdaq: TCOM, HK: 9961), Jianpu Technology (NYSE: JT) and a handful of domestic startups. A graduate of Fudan University, he writes widely about China's business and tech scenes and other topics for global publications including South China Morning Post, SupChina, The Diplomat, CGTN, Banking Technology, among others, and tries to impart his experience to students at Fudan University Journalism School, where he is a part-time lecturer. When he's not writing about robotics, you can expect him to be on his beloved Yanagisawa saxophones, trying to play some jazz riffs, often in vain and occasionally against the protests of an angry neighbor. Get in touch with him by dropping a line at nitao0927@gmail.com.

Articles: 662