Author: Li Mingzi
Published in the 1130th issue of "China News Weekly" magazine on March 4, 2024
At the Hefei NIO Automobile Factory, the industrial version of the humanoid robot Walker S from Shenzhen company Youbi is undergoing "practical training".
After checking the door locks, seat belts and light covers, the 1.7-meter-tall mechanical "employee" walked to the front of the car and very gently put the car logo on the new car.
On the other side of the ocean, in the United States, not long ago, the American start-up company Figure sent its first humanoid robot, Figure 01, to work at BMW's Spartanburg automobile factory, replacing humans in some highly dangerous jobs.
With the breakthrough progress of artificial intelligence, the humanoid robot as the best carrier will no longer be a "tool man" that can only deliver food and tighten screws.
Large models will empower humanoid robots to move toward active interaction, making them a disruptive product after computers, smartphones, and new energy vehicles.
Every step humanoid robots take can set off a market boom, and the pace of commercialization of humanoid robots is also accelerating in the competition.
China is also one of the most active surfers in this wave.
On October 20, 2023, the Ministry of Industry and Information Technology issued the "Guiding Opinions on the Innovation and Development of Humanoid Robots" (hereinafter referred to as the "Opinions"), pointing out that humanoid robots integrating artificial intelligence, high-end manufacturing, new materials and other advanced technologies will profoundly change human production and lifestyle, and reshape humanoid robots. Shaping the global industrial development pattern.
Currently, humanoid robot technology is evolving at an accelerated pace and has become a new highland for technological competition and a new engine for economic development.
Another national competition has begun.
The aforementioned "Opinions" of the Ministry of Industry and Information Technology proposed that by 2025, my country's humanoid robot innovation system will be initially established; by 2027, the technological innovation capabilities of humanoid robots will be significantly improved, and the comprehensive strength will reach the world's advanced level.
Rapid evolution
At the 2023 World Artificial Intelligence Conference last summer, Shanghai Fourier Intelligent Technology Co., Ltd., which specializes in rehabilitation robots, released its first general-purpose humanoid robot.
However, some artificial intelligence companies can't wait for the final version of the product. Zhou Bin, president of Fourier's Intelligent General Robot Division, once revealed to the media that "during the research and development process, companies have successively bought humanoid robots to verify their AI algorithms."
The humanoid robot industry is heating up with the emergence of large models.
The emergence of GhatGPT ignited the "Battle of Hundreds of Models" on a global scale, and embodied intelligence is considered "the next wave of artificial intelligence development". An intelligent system that can understand, reason and interact with the physical world requires a " Human body", humanoid robot is undoubtedly the best carrier.
"Introducing large models of artificial intelligence is equivalent to equipping humanoid robots with intelligent brains." Zhang Chi, chairman and founding partner of Beijing Xinding Rongsheng Capital Management Co., Ltd., told China News Weekly.
The development of AI has allowed the market to see the possibility of humanoid robots being implemented. Humanoid robots can understand and execute instructions and complete real work, and are no longer mascots that can only dance and show off.
Unlike traditional industrial robots that only exist to complete specific tasks, humanoid robots have been burdened with the expectation of intelligence since their birth - to experience the world, make decisions and perform tasks like humans.
In 1973, Japan's Waseda University launched WABOT-1, the world's first full-size humanoid robot. It can communicate with people in Japanese, measure direction and distance, walk slowly and grab objects. It is equivalent to the intelligence of a one-and-a-half-year-old child. .
"Japan is good at developing the physical characteristics of robots, while the United States is good at developing robot thinking." As Martial Herbert, a professor at the School of Computer Science at Carnegie Mellon University, summarized, the United States is leading in the field of artificial intelligence.
Interdisciplinary cooperation in computer science and mechanical engineering has allowed American companies to gradually replace Japan as the leader in the field of robotics. A landmark event was the robots involved in the accident handling after the nuclear leakage at Japan's Fukushima nuclear power plant due to the earthquake and tsunami in 2011. From the American company iRobot.
In order for humanoid robots to interact with the physical world, they must have the ability to perceive and understand, and these are inseparable from the development of AI.
"If deep learning allows artificial intelligence to perceive the world, the currently popular large models allow artificial intelligence to have autonomous decision-making capabilities." Zhang Chi said, for example, that the emergence of large AI models has given humanoid robots more powerful capabilities. In theory, as long as the robot learns enough data, it can have human-like intelligence, break away from the preset planning and make independent decisions.
At the same time, the establishment of "multi-modal" capabilities allows the robot to process information in multiple threads and realize the entire process of perception-decision-execution.
“AI gives this round of humanoid robot craze the imagination space to become universal.”
In fact, humanoid robots equipped with intelligent brains are evolving rapidly.
In Google Labs, the robot based on the "visual-language-action" model "RT-2" has been able to complete the instruction of "picking up extinct animals". On the table with three plastic toys of a dinosaur, a whale, and a lion, the robot Accurately picked up the dinosaur.
It can be deduced that as long as there are enough data, algorithms and computing power, robots will be able to handle more complex scenarios.
Late at night on January 4 this year, Google DeepMind released three new developments, all based on the RT-2 model, used to improve the speed, data collection and generalization capabilities of the robot.
Earlier in the day, a three-person team from Stanford University released a robot demonstration video based on the Mobile ALOHA system. The trained robot can complete complex mobile control tasks such as cooking, cleaning desktops, and pressing elevators.
In a demonstration video released by the American start-up company Figure, the robot Figure 01 can learn to make coffee through observation in just 10 hours.
In the first month of 2024, Stanford, Google, Figure, and Tesla successively released at least 6 new developments in humanoid robots.
"From academia to industry, this wave of enthusiasm for humanoid robots was largely driven by Tesla." Zhou Jian, chairman of the board of directors, executive director and chief executive of UBTECH, told China News Weekly that in 2022 , after Tesla moved the prototype of the humanoid robot Optimus Prime for display, the domestic industry responded quickly, and even the chairman of the manufacturer came to seek cooperation.
By the end of last year, in the demonstration video of the second generation of Optimus Prime released by Tesla, a dexterous hand with a tactile sensor and 11 degrees of freedom could already use two fingers to gently pick up an egg, and it seemed that more "delicates" could be accomplished. Tasks such as assembly, flexible object manipulation, etc.
Optimus Prime was last seen on Elon Musk's social media, showing the robot calmly folding clothes in a video.
According to Musk's vision, the robot's functions will be continuously updated, and in the next three to five years, a version will be launched that can be purchased by mass consumers.
For a long time, the biggest constraint on the commercialization of humanoid robots has been the high cost. Whether it is Boston Dynamics Atlas, which was once funded by the US Department of Defense, or Honda ASIMO, which has been discontinued, the cost is as high as millions of dollars.
"Tesla has provided a clear landing scenario for humanoid robots. More importantly, Musk said that he will reduce the cost to about 20,000 US dollars." Zhou Jian said, which allowed the industry to see large-scale mass production and commercialization. With the possibility of industrialization, "humanoid robots have reached the eve of industrialization."
This also makes many Chinese entrepreneurs eager to try.
The first batch of "pioneers"
For Zhou Jian, choosing a humanoid robot for his second business venture was far more difficult than originally imagined.
In 2008, Zhou Jian was attracted by the small and agile humanoid robot at an exhibition in Japan. Looking at the price of tens of thousands of yuan, he came up with the idea of "developing humanoid robots that ordinary families can afford."
At that time, industrial robots received widespread attention due to the development of manufacturing industry, but the industrial environment for service robots facing mass consumers was still very lacking.
The secret to making the robot's hands and feet move is the "servo drive", which is the "joint" of the robot. At that time, it could only be imported from South Korea, Japan, and Switzerland. Each joint cost hundreds of dollars, and it took more than a dozen joints to make a robot.
If a product wants to achieve mass production and commercialization, it must be independently developed and break through foreign technical barriers.
"The biggest difficulty is that there is no experience to follow." Zhou Jian recalled that in the early days of starting a business, the R&D team bought robots from various countries and brought them back for disassembly and training. Each link required countless trials, errors and improvements.
All the more than 20 million yuan in liquidity in Zhou Jian's hands was quickly burned out, like being thrown into a bottomless pit.
At that time, humanoid robots were an unpopular industry. They were not favored by venture capital and lacked market financial support. Zhou Jian gave himself a "blood transfusion" and invested all his savings for his first business. He sold houses and cars to subsidize research and development, and even borrowed money when he was most embarrassed. Pay wages.
After working hard on the research and development of servo drives for five years, the self-developed "joint" was not only comparable in quality to overseas first-line products, but also obtained multiple patents. The cost price was also reduced to a few tenths of that of imported parts.
The "stuck neck" problem of core components has been solved, laying the foundation for the subsequent commercialization of the Ubiselect robot.
In 2014, UBTECH’s first small humanoid robot, Alpha, was successfully mass-produced, and its price dropped from tens of thousands of yuan to several thousand yuan.
In order to improve R&D and production efficiency, Zhou Jian chose to establish a company in Shenzhen.
The Pearl River Delta region has more complete product supply facilities, making it easier to find factories that meet production requirements.
Zhou Jian once made a rough calculation that after the company settles in Shenzhen, the production efficiency of various parts will increase by at least 50%, and the cost will also be saved by 30%.
During a road show, UBTECH was spotted by Xia Zuoquan, co-founder of BYD and founder of Zhengxuan Investment, and received angel round financing, successfully solving the company's funding problem.
"I think our success in obtaining investment means that, in a sense, the market believes in Shenzhen's enterprise and industrial foundation." Zhou Jian said.
In Zhou Jian's view, the country's complete industrial manufacturing system and industrial chain provide guarantee for the mass production of humanoid robots.
Humanoid robots can share many similar supply chains, such as batteries, chips, sensors, controllers, etc., with industrial robots, new energy vehicles and other industries, enabling collaborative migration of supply chains.
After the successful mass production of his first small robot, Zhou Jian turned his attention to the development of more challenging large-scale humanoid robots.
Contrary to conventional wisdom, it is more difficult to make a robot walk on two legs than think with its "brain", and it is still a problem in the world.
At that time, there were many humanoid robot projects in domestic scientific research institutes, but the progress was relatively slow.
Zhang Rui, founder and CEO of Beijing Iron Man Technology Co., Ltd., once worked at the Institute of Unmanned Aerial Vehicles of Beihang University, the Institute of Automation of the Chinese Academy of Sciences, and the Fifth Academy of Aerospace.
He admitted that the majors studied in the institutional team were too close, the way of thinking was not broad enough, and they were easily affected by changes in plans.
Considering the pressure of final results inspection, some cutting-edge R&D projects with immature technologies are not easy to receive corresponding resource support.
The limelight of large-scale humanoid robots has always been occupied by overseas companies and research institutions.
Companies such as Boston Dynamics in the United States, Honda in Japan, and RobotCub Consortium in Italy are all spending money on the "humanoid robot" project.
In contrast, domestic R&D started late and is mainly project research-oriented.
It was not until 2000 that the National University of Defense Technology developed the first humanoid robot "pioneer" in China. By 2015, Beijing Iron Man Technology was established, and domestic "bipedal large humanoid robots" gradually came out of the laboratory and tested commercial waters.
"The biggest R&D difficulty is that the robot may fall at any time while walking." Zhang Rui explained that the core technical difficulty lies in the robot's motion brain composed of gait control, environmental perception and other links.
In order for the machine to achieve a dynamic balance during walking like a human being, it is necessary to design an algorithm separately according to the different states of the robot. The core step is to conduct mathematical modeling and analysis of the robot, and then calculate the mechanical structure based on the preset The weight, degree of freedom and other indicators are analyzed to calculate the torque and rotation speed of each joint.
To make the robot go from drawing to reality, engineers and factories need to repeatedly run in, assemble and test it.
In Zhou Jian’s view, there is a gap between domestic research and development of humanoid robot algorithms and the international advanced level. Some high-end sensors are still relatively weak. However, in the domestic core parts supply chain, some parts such as servo motors, control systems and reducers are It already has certain international competitiveness.
“It is only a matter of time before core components are domestically produced.”
The aforementioned "Opinions" of the Ministry of Industry and Information Technology also proposed that by 2025, a number of key technologies such as "brain, cerebellum, and limbs" will achieve breakthroughs to ensure the safe and effective supply of core components.
Encouraged by policies, Beijing established its first provincial-level humanoid robot innovation center in early November 2023. It plans to carry out research on five key tasks, including the prototype of a universal humanoid robot body and a universal large model of humanoid robots. The legal representative is Chief Executive Officer Youbixuan. Technical Officer Xiong Youjun is in charge, adopting a "joint attack" model to address the stuck neck problem.
"The center brings together important units in the upstream and downstream manufacturing chains of China's humanoid robots. In the future, we will work together to overcome difficulties in neck-stuck technology, including core components, motion control and other aspects, and promote industrialization," Zhou Jian said.
Commercial "asceticism"
On the morning of the last trading day in 2023, Zhou Jian, who has been in business for 16 years, led the 11-year-old Youbi to go public on the Hong Kong Stock Exchange. Accompanying him was the company's newly developed humanoid robot Walker S for industrial use.
Before the launch, UBTECH sold more than 760,000 robots.
Among them, the Walker series, a humanoid robot product that is truly close to the concept of "embodied intelligence", has sold only 10 units in two and a half years.
"The problem is that current humanoid robots are not smart enough. Even in the most promising scenario of the automobile production line, they cannot replace existing industrial robots, let alone people. Enterprises will not pay for 'big toys'. "Zhang Chi pointed out that large factories are basically monopolized by the four giants of international industrial robots. Germany's KUKA has core customers such as Mercedes-Benz and BMW in the automotive industry. Switzerland's ABB has outstanding motion control core technology advantages. Japan's FANUC and Yaskawa are in the automotive manufacturing industry. and electronic and electrical industries have extensive layout.
Instead of working in factories, family care and companionship are considered to be the most ideal scenarios for humanoid robots.
"But this is the long-term goal of industrial development." Zhang Rui said that, just like the problems faced by established industries, "robot nannies" cannot be mass-produced immediately due to intelligence and price reasons. To achieve true marketization, customer recognition and sales must be solved. System issues such as channels, laws and regulations, and ethics.
"The functions that everyone sees now, such as greeting and simple grabbing, can basically be realized on the third-generation humanoid robot in 2017." Zhang Rui introduced that Iron Man's humanoid robot has now been upgraded to the fifth generation and will be launched in 2021. Won the bid for the country’s first robot astronaut project.
To a certain extent, this realizes Zhang Rui’s original intention of starting a business.
In the past ten years, the United States and Russia have successively sent humanoid robots into space to perform scientific research, maintenance and other tasks on the space station.
Humans will experience muscle atrophy in the weightless environment of space for a long time, and robots will eventually become an important part of the history of aerospace technology development.
Zhang Rui proposed developing an astronaut robot in 2015, but there was no relevant project approval and funding at that time, so he decided to leave to start a business.
"Not all jobs and environments necessarily require humanoid robots." Zhang Rui also admitted that it is an inevitable trend for machines to make up for the shortage of labor, and will even replace people in some scenarios. However, the current market demand is not significant. Those with clear needs are mainly Individual fields such as university scientific research, aerospace and military.
The commercial exploration of robots has gone through a long detour.
Before this wave of industrial boom driven by big models and Musk, the domestic robot industry had already experienced two entrepreneurial peaks.
The first batch of local robot companies were established around the 1990s. Most of them eventually evolved into robot integrators or agents. There were few independent innovators. Even if companies later went public, their market value rarely exceeded 5 billion yuan.
After 2015, in line with the international "Industry 4.0 Strategy" and encouraged by favorable domestic policies, a large number of industrial robot companies emerged, but only a few persist to this day.
"It is difficult to start a business following the trend in the long run. Enterprises must ultimately solve practical problems and create new products." Zhang Rui said.
The well-known ASIMO was defined by Honda as the highest technical representative, but was eventually disbanded in 2018 due to its limited actual value to the company.
Zhang Rui once made a special trip to Japan to visit the Honda ASIMO team. The personnel have been dispersed to other departments, so that the technology accumulated in the development of humanoid robots can be reused in other products, because this is more in line with the development logic of a commercial company.
Zhang Rui chose a similar approach to Honda ASIMO.
When Iron Man Technology completed the research and development of the first three generations of robots and was preparing to enter commercialization, Zhang Rui made the largest organizational structure adjustment since the company was founded. The R&D team that was initially established based on technical division of labor was dynamically transferred to the company based on project needs. in different commercialization projects.
"Humanoid robots are the technological source of Iron Man Technology, but we never regard them as the only product. What really brings stable cash returns are more practical products that can meet customer needs." Zhang Rui gave an example, such as robots used in mountain transportation. Products such as unmanned vehicles, robots that can perform mining tasks in pipelines, and handling robots for smart factories are "currently only for corporate orders and not for mass consumers."
Once the humanoid robot is made, who will it be sold to?
"The humanoid robot industry lacks effective customers, and the country needs to guide several large industries to prioritize the use of humanoid robots." Zhang Rui believes that the demand-side force on industrial development often gets twice the result with half the effort.
Tesla’s Optimus Prime is considered the product closest to commercialization, thanks to the technology accumulation of autonomous driving in the “front-end industry”.
Through automobile research and development, Tesla has completed self-research on all core software and hardware from the cloud to the terminal, and has created a "robot with wheels."
Even so, Musk also said in 2023 that he did not expect that every component of humanoid robots would need to be re-developed, and he had to keep calm about the humanoid robot industry.
According to Zhangchi’s observation, in the field of humanoid robots, domestic software and hardware technology is still developing rapidly, but the current products are immature and the technical route has not yet been established, so it is still too early to talk about subsidies.
“What the industry really needs to think about is, who are the customers who can pay for humanoid robots?”
The test is “national investment”
When technology enthusiasts marveled at Optimus Prime's proficiency in housework, Musk quickly issued a "disclaimer" on social platforms.
He reminded that Optimus Prime is not yet able to fold clothes autonomously, but will definitely be able to act completely autonomously in any environment in the future.
In other words, when completing some complex actions, there may be a remote control person behind the robot, just like a trained game character.
The Mobile ALOHA project team has come forward to explain that for simple actions such as grabbing and placing objects, the Mobile ALOHA robot can achieve a 90% success rate after 50 times of learning, while for some complex tasks, such as loading plates and beating eggs, It still requires human intervention.
"Mobile ALOHA should be regarded as a well-made sports car, but it still needs a human driver to express its capabilities." Jim Fan, a senior scientist at NVIDIA, commented in an article.
At this stage, robots are not yet fully autonomous, and the "learning" process is child-like. Fancy rollovers are the norm. As the person in charge of the Mobile ALOHA project complained, "Robots are not ready to take over the world yet."
Between the omnipotent world of simulation and the ever-constrained reality, there is a barrier built by basic material science.
"Large models are currently unable to bring disruptive changes to the humanoid robot industry unless the gap between software and hardware can be bridged to unify mathematical models and physical models," Zhang Rui said.
"The conditions for large-scale industrialization of humanoid robots are not yet mature in the short term, and we need to be fully prepared for long-term investment." Zhang Rui said bluntly that the humanoid robot technology currently mastered by humans is lagging behind compared with the structure of the human body.
With the development of technology, the humanoid robot technology that has been used now and will be used soon will likely be eliminated.
"Long-term and stable R&D investment also requires the support of the national, local government and all sectors of society."
Since August 27 last year, regulatory authorities have continued to tighten IPOs, and there have been rumors in the market that "loss-making companies are not allowed to go public, and the fifth set of listing standards for the Science and Technology Innovation Board has been suspended."
However, hard technology companies may still be given the green light.
According to the 21st Century Business Herald, loss-making companies with hard technology attributes and growing operating income can still continue to promote the process of listing on the Science and Technology Innovation Board, provided that the market prospects are clear, the technology industry is leading, revenue has increased, and profits have declined due to heavy investment in scientific research. There are not many such companies that are either losing money or losing money, and the degree of loss is not high.
"Investors' exit prospects are not optimistic, so they naturally dare not invest in technology easily." Zhang Chi said that the fund investment cycle is generally 3 to 5 years, which is too short for an industry that is still in the laboratory stage and does not have enough time for industrialization. Enterprises The road to going public is tough.
There is no stable exit channel for venture capital funds. In the end, they can only pursue some industries with fast profits and a certain market, and dare not invest early in small-scale technology. As a result, technology projects will develop slowly due to lack of funds, and will fall into a new era. A vicious cycle.
“In the final analysis, to support technological innovation and industrial progress, we must solve the capitalization problem of funds.”
Venture capital's interest in humanoid robots has also begun to show new trends.
OpenAI, an artificial intelligence research laboratory, has also invested real money in humanoid robots.
1X Technologies, a Norwegian humanoid robot start-up, received Series A2 financing led by OpenAI Venture Fund in March 2023, amounting to US$23.5 million.
Galaxy Securities research report pointed out that 2024 is expected to be the first year of mass production of humanoid robots.
After the Ministry of Industry and Information Technology issued the "Guiding Opinions on the Innovation and Development of Humanoid Robots", Beijing, Shenzhen, and Shanghai successively issued industrial support policies.
Beijing has established a 10-billion-yuan robot industry fund. Shanghai plans to build a humanoid robot industry highland with global influence and build a "large model + humanoid robot" collaborative innovation platform. Shenzhen is relying on the "Shenzhen City to Accelerate the Promotion of High-Quality Artificial Intelligence" Action Plan for Developing High-level Applications (2023-2024)", which proposes to leverage the manufacturing advantages of the Guangdong-Hong Kong-Macao Greater Bay Area to carry out large-scale application of humanoid robots.
Xu Xiaolan, Vice Minister of the Ministry of Industry and Information Technology, once emphasized that in the next step, the Ministry of Industry and Information Technology will adhere to top-level design and promote the high-level development of China's humanoid robot technology and industry by guiding industrial innovation and development, overall machine promotion, software and hardware collaboration, and ecological construction.
At a time when population growth is slowing down and the aging rate is rising, the development trend of robots helping people solve key problems has become unstoppable.
Some voices even predict that at some point in the future, the ratio of robots to humans may exceed 1:1, and the number of robots will exceed humans.
"The commercialization of humanoid robots requires the joint catalysis of scenarios, technology and capital. At this moment, the human world may be in a period of profound change." Zhou Jian said.
"China News Weekly" Issue 8, 2024
Statement: The use of articles from China News Weekly must be authorized in writing.