2024, the first year of humanoid robots?
Latest update time：2024-07-11
Reads：
Text | Xiao Luyu
Source | Juchao WAVE
Humanoid robots have become one of the most eye-catching topics at the 2024 Artificial Intelligence Conference.
Humanoid robots from the National and Local Co-built Humanoid Robot Innovation Center, Dahua, Fudan University, Fourier, Yushu Technology, Kepler, Qingbao Robot, Leju and other companies and units formed an "Eighteen King Kong" array, demonstrating how to achieve dynamic coordination and complete unified movements through control. A large number of spectators stopped to watch.
Tesla's second-generation humanoid robot Optimus also made its debut in China, exhibited alongside Chinese-made robots. This has led many investors to excitedly declare that
"2024 is the first year of humanoid robots
.
"
Judging from the product appearance, the humanoid robots on display this year have made significant progress compared to last year, and the participation of more and more Chinese companies means that these robot products will be iterated and mass-produced more quickly.
However, the process of popularizing humanoid robots is still long, from laboratories to factories, from To B to To C, from booths to homes. It will take a long time for this largest smart device other than cars to enter the lives of ordinary people.
01
Scenes
The best application scenario is actually the family.
In recent years, many Chinese startups have launched their own humanoid robots, and their application scenarios are mainly focused on schools, industries, exhibition halls, etc., while only a few are suitable for home and medical use. This is mainly because the
commercialization of humanoid robots has gained some verification opportunities in industrial and commercial service scenarios.
Previously, Tesla Optimus was already the industry benchmark for humanoid robots. With the help of visual neural networks and FSD chips, the second-generation Optimus can imitate human operations and perform training such as gently holding eggs, carrying heavy objects, and sorting batteries. Its upright walking speed has also increased by 30%, and it has recently started working in Tesla factories.
Tesla stock price performance (from March 2024 to present)
Although Tesla has not yet exhibited the prototype offline and has only disclosed the dynamic effects of the second-generation Optimus in official videos, this has already made the market agitated - Tesla's stock price soared 27.11% last week, almost wiping out all the declines in the first half of the year, and the market expressed confidence in the narrative of humanoid robots.
Musk also called out Tesla shorts, saying, "Once Tesla completely solves the problem of autonomous driving and the robot Optimus begins mass production, anyone who still holds a short position will be eliminated, even Gates is no exception."
Tesla currently plans to achieve mass production and launch by the end of 2025.
By then, more than 1,000 second-generation Optimus will be deployed in Tesla factories to help employees complete production work. The longer-term goal is to achieve an annual output of 1 billion units in the future, occupy more than 10% of the market share, and bring Tesla 25-30 trillion US dollars in revenue.
Also going to work on the automobile production line are humanoid robots from UBTECH Robotics. Earlier this month, FAW-Volkswagen and UBTECH Robotics reached a cooperation agreement to introduce the industrial version of the humanoid robot Walker S to carry out tasks such as bolt tightening, parts installation, and parts transfer in the automobile manufacturing process.
UBTECH's industrial version of the humanoid robot Walker S will also be placed in the final assembly workshop of NIO's second advanced manufacturing base, Dongfeng Liuzhou Motor's final assembly workshop and other car factories for on-site "training". In the process, it will enhance its R&D capabilities, enrich the types of products available in industrial scenarios, and help factories improve their intelligence and unmanned levels.
In the short term, the application plans of various
manufacturers are concentrated in industrial manufacturing, warehousing and logistics, people's livelihood services and special applications. However,
in the medium and long term, the goal of humanoid robots is to enter thousands of households and provide relevant services for family scenes
.
The "Guanghua No. 1" displayed by Fudan University at this artificial intelligence conference is a nursing robot for the elderly. The robot is planned to be tested in Sichuan, Henan, Jiangsu, Zhejiang and other places to improve its ability to help the elderly get up from the bed and help them go to the toilet, and will try to carry out small-scale industrial promotion next year.
The XR4 "Xiao Zi" developed by Dahua Robotics has flexible hand size and can pick up eggs and playing cards. It can also complete delicate tasks such as toasting bread, serving plates, and ironing clothes. It is also suitable for multiple fields such as industrial manufacturing, home elderly care, and commercial services.
However, the pre-sale price of XR4 starting from 399,000 yuan per unit means that it is not so easy for this robot to move from the B-end market to the C-end market.
It can be said that the most suitable scenario for the application of humanoid robots is home service, and this is also the scenario that is most difficult to promote and apply at present.
02
intelligent
AI is completing the final piece of the puzzle of humanoid robots.
Most people expect humanoid robots to take on menial, unpaid housework rather than taking over paid jobs on the production line.
However, household services are actually more complex and open scenarios than industrial applications. Humanoid robots that can perform worthless housework well require much more R&D investment and capital support than ordinary industrial robots.
Without the explosion of large models, the future of household humanoid robots will be even more distant.
By learning from a large amount of text and images, large models provide robots with common sense knowledge, enabling them to handle various tasks and cope with complex scenarios, thereby enhancing the versatility of robots. Although there are doubts and debates surrounding whether to imitate humans or surpass humans, the application of large models is indeed driving a revolution in the field of humanoid robots.
Figure 01 is a good example of AI-enabled humanoid robots. In March of this year, 13 days after announcing its collaboration with OpenAI, the company showed a demonstration video of the humanoid robot Figure 01 equipped with the visual language model (VLM) provided by OpenAI.
In the video, Figure 01 can proactively pick out the only edible object from a pile of debris and hand it to a human. It can also recognize the status of the tabletop and determine that plates and cups need to be put back on the dish rack without specific instructions, and automatically execute its decision.
This means that Figure 01 achieves end-to-end processing of integrated perception, decision-making, and execution.
The control strategy adopts the OpenAI multimodal model, which significantly improves the intelligence level of embodied intelligence.
Integrating multiple senses such as vision, hearing, and touch, generatively coming up with solutions to problems and executing them in the physical world is exactly what a general humanoid robot needs. With the help of multimodal large models, all of this can begin to become a reality.
OpenAI has also recently restarted a robotics project that was disbanded four years ago to further promote the integration of artificial intelligence and robotics technology and explore new application scenarios and business models in order to maintain the company's leading position and competitive advantage in the industry.
As an "arms dealer" in the AI era, Nvidia is also trying to explore more business models and revenue sources for the implementation of AI by participating in the development of robotics technology, such as providing robotics software services, cloud platform services, etc., so as to make its company's valuation story more attractive.
NVIDIA has released a number of new products that support robots, such as the GR00T humanoid robot universal basic model that can receive multiple types of commands such as language, gestures, text, and cooperate with numerous software and hardware tools; the Issac Lab platform that can support simultaneous simulation training of thousands of robots; and the Jetson Thor humanoid robot computing platform based on the Thor chip that can provide 800 trillion 8-bit floating-point computing performance per second.
Nvidia stock price performance (January 2023 to present)
Companies that establish cooperation with NVIDIA are expected to benefit from the upgrade of AI computing platforms, simulation platforms, etc. NVIDIA also hopes to benefit from the wave brought by robots and autonomous artificial intelligence.
Of course, the challenges in the implementation of humanoid robots are not just about intelligence, but also include key issues such as power and energy management, mechanical design and motion control, perception and perception processing, intelligent decision-making and planning, human-computer interaction and safety, cost and availability.
A technological breakthrough may only require the inspiration of one or two geniuses, but cost control requires the efforts of the entire industry chain.
03
Cost reduction
Opportunities to make money in the industrial chain.
There is a saying in the industry that $20,000 is the cost threshold for large-scale promotion of humanoid robots. Tesla CEO Musk said that the future price of Optimus is expected to be less than $20,000 per unit.
However, the hardware cost of this product is currently about 50,000 to 60,000 US dollars, and its component value ranking is frameless torque motor (21%), reducer (16%), force sensor (16%), screw (14%), hollow cup motor (4%), inertial navigation imu (1%) and others (28%).
These key components have high technical barriers and are not only the main link in reducing costs for humanoid robots, but also an important direction for domestic substitution.
Taking reducers as an example, this is an area where the possibility of domestic substitution is relatively high. Green Harmonic's global sales market share is 7% (Japan's Harmonic has a global sales market share of 82%, and the sales market share of other manufacturers is about 11%). It has caught up with the international advanced level in terms of reduction ratio, rated torque, etc., breaking the technological monopoly of foreign manufacturers.
In order to prepare production capacity for the era of robots, Green Harmonic's 2 billion yuan private placement project has been approved recently. Before the annual production capacity of 1 million new-generation harmonic reducers and 200,000 mechatronic products is reached, it has already started a price war, reducing the price of products supplied to collaborative robot customers by about 20%.
This production capacity + price strategy not only affects the performance of Green Harmonic, but also has a wide-ranging impact on the overall ecology of the reducer industry
.
It reflects the determination of supply chain companies to bet on the huge growth of humanoid robots amid sluggish demand for consumer electronics
.
Screws are an area where domestic manufacturers are catching up. Japan's THK, NSK and Germany's Rexroth and Schaeffler occupy a major share of the mid-to-high-end market. In order to cater to the technological wave of humanoid robots, companies such as Wuzhou Xinchun, Best, Beite Technology, and Hengli Hydraulics are also gradually entering the high-precision ball/roller screw market.
At present, the main barrier of humanoid robot lead screws lies in the fine processing of internal threads. If technological breakthroughs and mass production can be achieved, robot manufacturers can shift their focus to large-scale cost reduction. Although this will place high demands on the cost control ability and yield rate of supply chain companies, it also brings opportunities for domestic substitution.
In addition to core components, processing equipment such as grinders, lathes, and winding machines, new technical directions related to material cost reduction and lightweighting, such as PEEK materials, ductile iron technology, and cold forging technology, are also expected to get a share of the humanoid robot craze.
Chinese humanoid robots have hardware production advantages
Although Chinese companies still need to focus on Tesla for the time being, they have already begun to catch up in technology. Currently, the total number of Chinese humanoid robot technology patent applications exceeds 6,000, surpassing Japan, South Korea, the United States and other countries, ranking first in the world. In terms of valid invention patents, it also ranks second in the world with nearly 1,700.
Compared with technological breakthroughs, mass production and commercialization may be more difficult to solve. Cost and pricing issues are just the explicit expression of the former. Therefore, manufacturers are not pursuing mass production, but gradually promoting intelligent improvement and application scenarios, welcoming the advent of industry blockbuster products in iterations, thus opening up real commercialization.
It can be said that the biggest cost for humanoid robot companies is actually time.
-END-
The content of this article is for communication and learning purposes only and does not constitute any investment advice. The cover image is from Photo Network. If you have any questions, please contact us at info@gsi24.com.
▼
Review of past highlights
▼
Net profit soared by more than 754%, Ningbo's richest man in the chip industry made a lot of money
How do automotive MCU companies stand out from the competition?
The invincible electronic "cockroach"! This chip has been a best-seller for 50 years
After spending 840 billion yuan, Samsung was stabbed in the back by its own people
The chip leader's Q2 profit increased by 1452%. What's the reason?
京公网安备 11010802033920号