Humanoid robot track surprises continue to surprise Why is another "iPhone moment" coming?
"Science and Technology Innovation Board Daily" on October 2 (editor Song Ziqiao) Looking back on the past three quarters, the humanoid robot product side has many highlights, once again refreshing everyone's cognition. The most important event is that Tesla's humanoid robot quickly iterated and showed off the operation again.
On September 24, Tesla's humanoid machine Optimus demonstrated its ability to sort items autonomously and maintain balance on one foot through a video. Tesla said Optimus' neural network training is "completely end-to-end" — video signal input and control signal output without preset code.
The near-complete Optimus was unveiled in September 2022, and viewers were able to catch a glimpse of Tesla's second AI Day, where it was not walked around the site and was carried by several big men throughout the process. Since then, Tesla has been broadcasting its latest progress in the form of a video. Video from March 2023 shows Optimus already able to walk; In the May 2023 video, Optimus can perform simple tasks.
Why did the progress this time exceed expectations?
Unlike the previous two videos this year, in addition to the visual perception and motor control upgrades, Optimus focuses on its human-like learning mechanism - video input, control output, that is, learn a skill by seeing, listening, and thinking, rather than relying on code entered in advance. This learning mode is the basis for the robot to adapt to the changing environment, as long as enough human behavior data is collected, the robot can quickly imitate and learn the corresponding behavior.
This mode is similar to Tesla's automatic driving training mode, which once again proves that Tesla's successful experience in FSD and Dojo (including underlying modules such as algorithms) is expected to be reused in robot training, thereby accelerating the iteration of humanoid robots.
A smart electric car is essentially a four-wheeled robot. Tesla's electric vehicle has a pulling effect on the local automotive industry chain for all to see, and now, this catfish effect seems to be re-emerging in the robot industry.
While Tesla robots show off their muscles, "Huawei Genius Boy" Zhihuijun's "Expedition A1", UBTECH's Walker X/Panda Robot Youyou and other domestic humanoid robots have also appeared, and Fourier Intelligent General Humanoid Robot GR-1 also opened pre-sale on September 26.
The breakthrough of the product side is only one dimension of industrial development, and favorable policies and technological progress are jointly promoting the commercialization of humanoid robots. China is already the world's largest robot consumer market, and the question that arises for domestic investors is: where are the opportunities for the local industrial chain in the new round of robot innovation cycle?
▌Investment opportunities for local enterprises: hardware side> software side
The humanoid robot industry chain mainly includes the software side (environment perception module, intelligent AI chip module, operating system module) and the hardware side (motion control module).
At present, empowering robots to act is still the most difficult problem in the engineering field, and this link is also regarded as an opportunity for the local industrial chain. Based on the views of many institutions, for local manufacturers, whether it is industrial robots or humanoid robots, the value of motion modules on the hardware side is the highest.
Sun Xiaoya, an analyst at Tianfeng Securities, took Tesla robots as an example, saying that the software side, especially the algorithm related to the high probability of Tesla doing it itself, so from the perspective of A-share investment opportunities, domestic hardware suppliers have greater opportunities.
CITIC Securities analyst Yuan Jiancong also said that OpenAI, Tencent, Xiaomi, Google, NVIDIA and other technology companies continue to inject in the field of robotics, promote robot technology innovation and reduce costs and mass production process accelerated, robots are currently in the early stage of development, there is a "hardware first" law.
For humanoid robots, the three core parts of the hardware part are rotary joints, linear joints, and hand joints, and the corresponding core components are motors (frameless torque motors, coreless motors), harmonic reducers, lead screws (planetary rollers, T-type sliding), sensors, etc.
Taking Optimus as an example, many institutions expect that its hands mainly use hollow cup motors, leg rotary joints mainly use frameless torque motors + harmonic reducers, and leg linear joints mainly use frameless torque motors + planetary roller screws. Its motion module and part of the perception module are integrated into an integrated joint, which accounts for up to 50% of the total cost.
▌Parts suppliers are eager to try, how far is commercialization?
According to incomplete statistics from the "Science and Technology Innovation Board Daily", in the third quarter, more than 100 A-share companies were asked about the business progress of humanoid robots/embodied intelligence/robots on Interactive Easy, including not only robot/robot parts manufacturing enterprises, but also medical device companies, lithium battery manufacturers and other enterprises that may apply robots.
Among them, Rujing Technology said that its servo drive and control system products have achieved mass sales in industrial robots and other sub-industries; Jiangsu Reilly said that at present, the company is in orderly contact with a number of robot manufacturers, and the sample testing stage is progressing smoothly; Topstar says a small number of its industrial robots are used in Huawei production lines...
According to the statistics of the "Science and Technology Innovation Board Daily", since July, dozens of A-share robot companies have been investigated by institutions, and upstream hardware suppliers have received higher attention than complete machine manufacturers, and the number of companies investigated is more, including Haozhi Electromechanical, KONE Technology, Fengli Intelligence, Buke Co., Ltd., Jiangsu Reilly, Zhongdaide, Xinshida, Southern Precision, Precision Forging Technology, Wuzhou Xinchun, etc.
Looking through the research minutes of these companies, in the institutional questions, the "content" of parts is extremely high, and terms such as torque sensor, harmonic reducer, RV reducer, planetary reducer, coreless motor, frameless torque motor, planetary roller screw, sliding screw and so on have repeatedly jumped into the eye.
However, only a few companies such as Jiangsu Rayleigh, Zhongdaide, and Topband have responded to the humanoid robot business.
Domestically, robots have been listed as key development projects in Beijing, Shanghai and Shenzhen, and industrial support policies have been issued successively.
At the same time, the cooperation and research layer of domestic industry and academia is endless, such as Boshi Co., Ltd. and Harbin Institute of Technology to set up a humanoid robot key technology and principle prototype industrialization research and development project, Tianqi Co., Ltd. and UBTECH to set up a humanoid robot joint venture. AI technologies such as GPT large models have become a booster, and many researchers have tried to use multimodal large language models as a bridge between humans and robots.
Looking forward to the future, after the commercialization of humanoid robots, the product side is expected to usher in a situation of a hundred flowers, how far are we from this moment?
A number of institutional analysts take Tesla as a weather vane, China Galaxy Rupe believes that the export of Tesla robots will be a landmark event in the industry, the importance of analogy to the release of iPhone 1 to open the era of smartphones, Tesla Model 3 sales to open the era of new energy vehicles.
Tesla's new round of "road show" has been put on the agenda. At Tesla's second-quarter earnings briefing on July 20, Musk expected Optimus to conduct walking tests in November and utility tests at Tesla's own factories next year.
(Science and Technology Innovation Board Daily, Song Ziqiao)