news

"we started small-scale trials and deliveries at the end of last year, and have now delivered more than 100 units. this number is a major milestone for the humanoid robot industry," said gu jie, founder and ceo of shanghai fourier intelligent technology co., ltd. (fourier), in a recent interview.

the more than 100 humanoid robots mentioned by gu jie are mainly the fourier gr-1 humanoid robots, which were released in july last year, pre-sold in september, and delivered in october.

one year later, on september 26, 2024, fourier's independently developed new generation universal humanoid robot gr-2 was officially released to the public. "we collected a lot of customer feedback, summarized thousands of improvement suggestions, condensed these suggestions bit by bit, and iterated and upgraded to give birth to the second generation of robots," said gu jie.

the gr-2 released this time is 175cm tall and weighs 63kg. it has 53 degrees of freedom in the whole body and a single arm movement load of 3kg. it can complete more complex operations than the previous generation gr-1. the battery capacity of gr-2 has doubled, and the battery life has been extended to 2 hours. in addition to supporting direct charging, a detachable battery replacement solution has been added to fully meet sports needs.

"the fact that humanoid robots can be delivered to the outside world in small batches means that we don't just keep the robots in the laboratory. we have the courage to put them into real scenarios and let them be tempered by users and scenarios. only after delivering hundreds of them can we move to thousands and then tens of thousands, and realize the grander vision of having robots around us," said gu jie.

in gu jie's view, the vision of "empowering life with robot technology" was still far away five or ten years ago. "when the company announced that it would make humanoid robots, it was april fool's day, and there were only a few companies making humanoid robots in the world." at that time, there were no large models or powerful power units, but ten years have passed and the industry has changed dramatically, especially with the emergence of gpt in the past two years. the original robots were fixedly programmed, but now they have the ability to reason, think and communicate; the original robots were very dangerous and had to be fenced off, but now they can be right next to you. "the speed of technological development is very fast!"

gu jie's emotion and the rapid iteration of fourier humanoid robots are actually an important epitome of the development of the embodied intelligence industry in the past two years.

looking at all the major offline exhibitions in the past two years, intelligent robots have always been an important highlight. at the 2024 global artificial intelligence conference in shanghai, a row of "eighteen arhats" composed of humanoid robots in the exhibition hall was once surrounded by the audience; at the recent yunqi conference in hangzhou, various robots also "took turns to the stage" in one exhibition hall...

at the same time, industry forums and discussions on embodied intelligence have become standard, especially on how large models can enhance the capabilities of robots.

wu yongming, ceo of alibaba group and chairman and ceo of alibaba cloud intelligence group, also said at the recent yunqi conference: "robots will be the next industry to undergo a huge change. in the future, all movable objects will become intelligent robots. they can be robotic arms in factories, cranes on construction sites, porters in warehouses, firefighters at fire scenes, and even pet dogs, nannies, and assistants in families."

36kr research institute also recently published a report stating that humanoid robots have made significant progress in industrial applications, and the demand for applications in manufacturing, new energy, medical and other fields has continued to rise. many companies have released humanoid robot products or iterative versions and achieved technological breakthroughs. however, they are still facing scale challenges, which are mainly attributed to the high technical difficulty, high manufacturing costs and high commercialization difficulties. the industry predicts that as technical bottlenecks are overcome, the industrial chain is improved and costs are reduced, the humanoid robot market is expected to grow rapidly, and the chinese market size is expected to reach 300 billion yuan in 2035.

building a good "body" for ai

"fourier's positioning is to create a true ontology platform and hardware platform for ai." gu jie, founder and ceo of shanghai fourier intelligent technology co., ltd. (fourier), emphasized in an interview that the company's humanoid robots must be made for ai and born for ai.

in gu jie's opinion, the large model of visual language has now been generated. if you give it a sentence, it can generate a video. if you give it a picture, it can know the specific description of the picture and has a visual understanding of the world.

“but this is not enough, our robots need the last link, to be able to generate actions. we need to predict not only the next word, but also the next action, so the big model in the future will be a big model of vision, language and action, that is, the vla model of vision, language and action. "gu jie said that the world has the strongest brains and the most advanced laboratories working on the vla model, and this model must be loaded on a body, and this body is a humanoid robot.

fourier wanted to build a good body and then load the best large model on it, so that it could be aimed at colleges and universities, leading enterprises, scenario developers, and end customers, and ultimately achieve task output.

in view of the future evolution of humanoid robots, gu jie proposed the concept of "product hexagon", which focuses onmotion intelligence, dexterous operation, cognitive intelligence, bionic design, user experience and commercial applicationthere are six dimensions in total, which clarify the capability standards and development direction of the robot itself.

"when the humanoid robot body reaches the optimal state in all dimensions and becomes a hexagonal warrior, it will definitely be the best embodied choice for ai. the ai ​​here is not limited to large models, but also includes the entire process of robot perception, decision-making, and execution." said gu jie.

many technologies still need to be broken through

although the discussion about embodied intelligence in 2024 has remained high, from the perspective of experience, there have been constant doubts. are humanoid robots a bubble? are they just in the demo stage and too far away from reality? is it still too early in the race? it seems that the product can only go a few steps and is not going well...

"we don't shy away from these issues, because fourier didn't start doing this just because there was an industry hotspot. we started wanting to do these things ten years ago." gu jie believes that in the past few years, some problems have been solved, while some have not yet been solved. "but for a technology, don't overestimate its progress in half a year or three months, and definitely don't underestimate the changes it can produce in three to five years."

in gu jie's view, the development trends of humanoid robots includetechnological breakthroughs, mass production, and scenario applicationsthree parts. for example, technological breakthroughs may come from all aspects, such as the most cutting-edge large models, breakthroughs in ai technology, breakthroughs in core components of the hardware itself...

gu jie divides the technical sector into the algorithm layer and the ontology layer. the algorithm layer is divided into the brain and the cerebellum. the brain is responsible for everything from perception, interaction to decision-making. it collects visual, tactile, auditory and force signals through various sensors, processes them and inputs them into the brain, which generates an action decision through large model calculations.

if you want to walk fast, march in place, or maintain balance, these movements do not require the brain to control the hip and knee joints. they are very smooth, conditioned reflexes, and instinctive movements determined by the cerebellum. the main body layer includes the whole body integrated design and key components, such as bionic dexterous hands, sensors, and joints.

the most core component is the integrated actuator, which is the joint of the robot. it includes a set of complex devices such as servo system, reducer, sensor, chip, etc. "what fourier has to do is to make the cerebellum and the body well." gu jie said.

after the technological breakthrough, we have a prototype. the second step is mass production and manufacturing, which involves a lot of manufacturability, processability, stability, manufacturing parameters, etc. it is a massive engineering problem. after manufacturing, it will eventually be put into use in scenarios. in the future, it may be used in thousands of industries.

it's still a marathon process

"when boston dynamics' hydraulic version of the atlas robot was launched in 2013, it cost about $1 to $2 million." gu jie said that in april 2024, boston dynamics announced the official retirement of the hydraulic version of the atlas robot and launched a new electric version of the atlas robot. this new atlas robot is completely battery-powered and no longer uses a hydraulic system, providing greater strength and a wider range of motion than ever before.

this is just like the automobile industry. new energy vehicles have become popular because they have changed from gasoline vehicles in the internal combustion engine era to electric vehicles. this is a major turning point, which has transformed it from a traditional manufacturing industry to an industry with consumer attributes and rapid iteration. humanoid robots are also experiencing such a huge change. many of the autonomous driving technologies of new energy vehicles can be transplanted to humanoid robots. the boundaries between products have become blurred, and the iteration speed is getting faster and faster.

"at present, the price of humanoid robots is about 1 million yuan. the price will definitely continue to drop in the future, but it is not yet the stage of a price war."

at tesla's shareholder meeting in june 2024, musk revealed that tesla will deploy thousands of autonomous "optimus" (humanoid robots) to work in super factories in 2025. by 2026, optimus robots are expected to reach a certain scale and begin to generate significant economic benefits.

musk expects that after the development of humanoid robots enters the mature stage, the ratio of the number of robots to the human population will be "at least 1:1, and likely 2:1 or 3:1."

in the latest interview, regarding the market pricing of the optimus humanoid robot, musk said that the price of the third-generation optimus will drop to about $10,000 in the future, and it will become a small robot companion that the public can afford. this robot can not only take on simple housework, but also accompany family life, becoming a good helper in human life like the characters in science fiction movies.

it sounds like musk is quite optimistic about humanoid robots. in the view of gu jie of fourier, the development of humanoid robots is still a marathon process that requires long-term investment.

"if humanoid robots are to be made, funding is a factor, but it is not a product that can be transformed by massive funding in one or two months. it requires the definition of the industry, continuous iteration, and continuous improvement." gu jie said that this involves technology and strategy, and there are too many scenarios to choose from. "from power motor control, mechanical design, software algorithm control, etc., it is a very large technology carrier. no company can complete and refine it all, and it needs to be more open."

in gu jie's view, today more and more young people and engineers are joining the humanoid robot industry. globally, one can vaguely feel that some efficiency of investment has begun to emerge, but this industry has just begun, and there may be a reshuffle in the structure in the next five or ten years.