한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

Text/ AI Future Guide by Zhou Xiaoyan

Editor/ Hao Boyang

This year, the humanoid robot track is very "hot", with constant news of financing, but it is actually a "false fire". Beneath the lively surface, there is no product with real generalization capabilities, a lack of customers, no sales, and no real mass production. The supply chain under the iceberg is like a "castle in the air".

Only some governments or institutions place small orders from enterprises for scientific research or scenario training purposes. These lightweight "mass production" are enough to make the top players represented by Musk distressed.

As a "madman", Musk always has some crazy ideas. From SpaceX, Neuralink, to Tesla, Musk has been tirelessly trying many challenges that ordinary people can't even dream of. Even if these things seem difficult, we rarely hear him talk about the difficulties behind them in public, but the humanoid robot supply chain is an exception.

At the beginning of 2021, Musk mentioned at a senior management meeting at Tesla headquarters that "we should seriously build a robot", which opened up a new challenge. Tesla is closely planning the manufacture of the humanoid robot Optimus.Musk originally thought that this would be a similar path to car manufacturing, after all, Tesla's successful experience was there. He once thought that making robots was just a "car without wheels", but he soon encountered a problem that he himself had not expected - the humanoid robot supply chain.

In a public video, Musk said helplessly, "Although there are many motor suppliers in the world, no motor is suitable for humanoid robots, and no gearbox meets our size requirements."

Brett Adcock, founder of Figure01, is in the same predicament. He lamented on social media X: “The biggest shock to me is that the humanoid robot supply chain is so immature. To be a humanoid robot company, you not only have to build robots, but also redesign parts.”

With no outstanding products and insufficient upstream supply of key components, every step in the development of humanoid robots is extremely difficult.

This kind of dilemma is not unique to humanoid robots. From early car manufacturers to today's smart device development, the birth of every "new species" is accompanied by product construction and supply chain problems. But history tells us that behind every problem, there are countless failures and attempts.

The relationship between products and supply chains is like the chicken and the egg. How can we break the deadlock? Perhaps history has already given us the answer.

01 No skillful cook can cook without rice

Looking at Musk's X account, you will find that since the decision to produce Optimus in 2021, a keyword has frequently appeared on his account: supply chain. However, the difficulties faced by humanoid robots are completely different from those faced by Musk in the past.Not only is there no "rice" to put into the pot, but due to the immature product technology and lack of practical application scenarios, there is not even a "skilled woman" to cook.

Therefore, if we break it down, the dilemma of the humanoid robot supply chain is reflected in two aspects. First, there is no robot product with real generalization capabilities; second, there is no "rice" to cook. In other words, even if you want to mass-produce in small quantities, there are no suppliers who can produce suitable parts.

In every era, there will be new species of hardware. There have also been fields like humanoid robots that build products from 0 to 1, such as Musk’s car manufacturing, but can this experience be reused on humanoid robots?

In 2007, the 36-year-old Musk was exploring the thorny issues in Roadster production. At that time, Tesla had not yet conducted a comprehensive review of the entire supply chain system. After ordering people to sort out the list of materials, Musk couldn't help but sweat. According to the description in "The Biography of Elon Musk", the battery pack in a Tesla car may have traveled around the world before reaching the hands of consumers: first the battery cells are produced in Japan, then assembled in a temporary factory in the jungle of Thailand, then transported to the UK by sea, and finally sent to the Lotus factory to complete the chassis assembly of the sports car.

During the subsequent running-in process, Musk gradually planned a supply chain path for Tesla: vertically integrate the supply chain, produce key components by itself, rather than purchasing hundreds of components from different independent suppliers and then assembling them into cars, and strictly control the quality of key components.

Looking back on his history several years later, the situation Musk faced in the car-making business can be summed up as “a good cook cannot cook without rice” - there is technology, products, potential customers and sales, but there is no supply chain and no “rice” to cook.

But humanoid robots face even more severe problems. Roadster is a To C product. "If it is feasible in the general direction, there will always be people willing to pay for To C products. There are 8 billion people in this world, and you can always find 8,000 or 10,000 consumers willing to pay," Deng Xigeng, an investor in the robotics track of Kunzhong Capital, told Tencent Technology. "But robots don't follow this logic. They are a 2B business. It's not as easy as a 2C business to find enough angel customers to get started and drive the maturity of the supply chain."

Wang Xingxing, founder and CEO of Yushu Technology, once told Tencent Technology that "the hardware for making humanoid robots is not the bottleneck. What currently limits the development of embodied intelligence as a whole is the level of AI." He believes that if a robot that can do the same things as humans really appears, "even if it sells for 100 million yuan, there will still be many people buying it."Therefore, in essence, humanoid robots are facing not only supply chain problems, but also the fact that their products and technologies have not yet been generalized to the point where enough customers will pay for them.“No company can make a robot that customers will want to buy 100 of.”

02 The "chicken and egg" vicious circle that cannot be broken

For humanoid robots at this stage, they are still in the research and development stage of "from 0 to 1", and are not in a hurry to think about the "beautiful trouble" of mass production such as "from 1 to 100". However, in order to improve the adaptability of humanoid robots in multiple scenarios and obtain the ability of "embodied intelligence", multiple robots must be produced for training in different scenarios.

Therefore, humanoid robots may have the trouble of lightweight mass production "from 0.5 to 10", which means that humanoid robots are trapped in the vicious circle of "which came first, the chicken or the egg". Without a certain level of mass production capability, it will be difficult to obtain a mature robot product through multi-scenario training; and if the robot product is not mature enough, it will affect further mass production.

So Musk complained about the lack of supply chain for humanoid robots, but at the same time he was worried about the supply chain, because Tesla could digest Optimus robots internally, which required Musk to produce a certain amount of Optimus robots. At the Tesla Investor Day event in 2023, a video clip of Optimus "making itself" was shown. From this perspective, compared with many founders of robot companies, Musk is already at the stage of "from 0.5 to 10".

Amazon and other robotics companies have invested $150 million in Agility Robotics, which has also started mass production. A year ago, Agility Robotics established the Robofab factory in Salem, Oregon, and plans to produce 10,000 Digit robots per year. Amazon has been testing the use of Digit in warehouse work. Digit can sense, grab and move large items such as cargo boxes, and scan the surrounding environment through the light detection LIDAR system to navigate the environment according to human design.

"But if we change the warehouse, Digit may not work," Deng Xigeng believes that although Digit has some commercial scenarios, its actual flexibility is very limited and it is still some distance away from mass production.

Therefore, at this stage, the "mass production" of humanoid robots is mainly concentrated in early industrial testing and scientific research needs, rather than real large-scale commercial needs. The number of units purchased for this type of demand may be in the hundreds.

For these two types of needs,It would be more advantageous for the host manufacturers of key components of humanoid robots to research the robot itself.For example, Tesla can provide Optimus with naturally rich training scenarios. In addition, in terms of key component suppliers, Tesla can design the robot's motors, gearboxes, and sensors from 0 to 1 based on the existing automotive supply chain.

This advantage is not only reflected in the rich training scenarios, but also in the independent control of non-standard core components. If we compare it to humans, the movements of humanoid robots are mainly divided into two categories: linear movement and rotational movement. There are three types of joints and related components involved. These three types of joints are mainly composed of three core components: motor, reducer and screw:

According to the report "Comprehensive Analysis of the Humanoid Robot Industry Chain", taking Tesla Optimus Gen2 as an example, the cost of the reducer + motor + screw accounts for more than 50% of the component cost.

Just one motor component involves upstream raw materials such as rare earth magnets and silicon steel sheets, midstream motor manufacturers and downstream humanoid robot companies. Different application scenarios require different types of motors. Humanoid robots pay more attention to the lightweight and dexterity of motors, so there are higher requirements for the compact structure of motors.

If there is only a small amount of order demand, it is almost difficult to leverage upstream and downstream to produce customized motors. At this stage, the humanoid robot supply chain is more suitable to adopt Musk's car-making approach, using a vertically integrated supply chain, combining original design, production and procurement from 0 to 1.

In the future, if the product is mature enough and reaches the "from 1 to 100" stage, the humanoid robot supply chain can refer to Xiaomi's start-up methods.When Musk was working on Tesla, there was no mature supply chain system or production standards in the world that matched the mass production of electric vehicles. In 2010, when Lei Jun was working on Xiaomi mobile phones, the mobile phone supply chain was already relatively mature, but in the environment at the time, leading suppliers were not optimistic about the concept of "Internet phones". In their minds, Xiaomi's mobile phone business was just a small business.

In order to attract the best suppliers, Lei Jun needed to knock on doors and visit each one to build his own supply chain. At that time, many suppliers suffered from "copycat phones" and were very wary of new mobile phone brands. Although some mobile phone brands had orders of hundreds of thousands, sales could not keep up, and suppliers faced indefinite delays and uncollected final payments. For Xiaomi at that time, "even the simplest screws, manufacturers were unwilling to cooperate with Xiaomi."

In the book "Forward" which describes Xiaomi's entrepreneurial journey, the author mentioned that "when a hardware company's turnover is around 10 billion, basically the global 3C (compulsory product certification system) manufacturing supply chain system is at your disposal." In the world of hardware, time is the cost of real money, and nothing is free.

The high production cost of humanoid robots cannot withstand the delay of time. A blogger named "Yang Moyu from the Investment Research Department" in his popular science video did some calculations using the Tesla robot as an example. The total cost of the three core components, rotary joints, linear joints and dexterous hands, which account for more than 50% of the hardware cost of humanoid robots, is about 260,000 yuan. The cost of a single humanoid robot is as high as 500,000 yuan (68,900 US dollars). This is in the same range as Morgan Stanley's recent analysis report "Humanoids: Optimus Prime", which predicts that the current cost of Optimus Gen2 hardware is between 50,000 and 60,000 US dollars.

In Musk's vision, the ideal price of Optimus is $20,000. The gap still needs to be filled through mass production. Morgan Stanley also gave a positive forecast, "With the expansion of scale, shortened R&D cycle and lower prices of Chinese parts, the cost may drop to the $20,000 target set by Elon."

Therefore, mass production is urgent, but is still limited by the maturity of the product.

Conclusion:

Humans can build cars, rockets, and robot dogs. These mindless anthropomorphic species seem to be easier to process with various materials, but they just need to repeatedly apply physical knowledge. But "humans" are a masterpiece of nature, and the parts of their bodies cooperate with each other in an extremely exquisite way. Humans have not even studied the precision of their own systems, but they are trying to create a species that is infinitely close to themselves. The complexity of the software and hardware involved in such a super species can be seen, and the complexity of its supply chain creation is far greater than that of other hardware.

If it’s so complicated, why rush to conquer it?

There is a question on Quora: Why does Musk want to make a humanoid robot?

There is a highly-praised answer that may answer both questions at the same time:

If a person has an interest in something and has unlimited funds available to them, they can create something, which, if nothing else, satisfies the urge to prove that they can do something that others cannot.

Perhaps, the group of people who make humanoid robots are a group of people with a desire to conquer. Although the road to "making humanoid robots" is full of ups and downs, they still overcome all obstacles.

"I am running towards you, and you are the sea of ​​stars."

material:

Yang Moyu from the Investment Research Department, "The simplest and clearest on the Internet! 20 minutes of hardcore disassembly of the core components of humanoid robots"

Morgan Stanley，《Humanoids:Optimus Prime》

Musk's annual speech at the 2024 Tesla shareholder meeting