The picture shows the engineers of GAC's third-generation embodied intelligent humanoid robot team recently explaining the breakthrough of AI autonomous control technology and application scenario innovation of this type of robot. Photo by Xinhua News Agency reporter Liu Dawei

Recently, the humanoid robot track has continued to heat up, and many car companies such as GAC Group, Xiaopeng Motors, and BYD have accelerated the layout of this frontier field. Industry experts pointed out that there are many fits between humanoid robots and the automotive industry, and with the advancement of technological breakthroughs and scenario verification, humanoid robots are expected to take the lead in large-scale application in industrial fields such as automobiles, and become an important engine for new productivity, but industrial development still needs to break through the bottlenecks of data collection, limited training scenarios, and underlying application ecological construction.

Car companies compete for the humanoid robot track

In the automobile production plant, many humanoid robots are busy on the production line. Some are working closely with the self-driving forklift, picking up the cargo box with both hands and steadily walking to the forklift to put it down. Some stand in a row in front of the shelves, and flexibly pick up parts of different shapes with ten fingers, and accurately classify and place them. In the face of heavier and larger goods, even if one "person" cannot move, other robots can be called to work together to lift them...... These sounding sci-fi movie scenarios are becoming a reality on the production lines of many car companies recently.

The reporter learned that with the continuous development of humanoid robot technology, many car companies are also accelerating the layout of this track. Some car companies choose independent research and development and take the lead in conducting practical training on their own production lines, while some car companies choose to cooperate with relatively mature humanoid robot companies to open production lines to provide training scenarios, and promote the in-depth implementation of humanoid robots in automobile manufacturing and industrial fields through real-life training and application exploration.

Zhu Jiangming, founder, chairman and CEO of Leapmotor, revealed that Leapmotor has set up a special robot team for technical pre-research. "In the future, Leaprun's robot products will mainly be used in the industrial field and will not be oriented to the C-end."

GAC Group has previously released the third-generation embodied intelligent humanoid robot GoMate, and announced that it plans to achieve mass production of self-developed parts in 2026 years to meet the growing market demand, and take the lead in the workshop production lines and industrial parks of GAC Trumpchi, Aion and other OEMs to carry out demonstration applications. In 0 years, the whole machine will be produced in small batches, and gradually expanded to large-scale mass production.

Xpeng Motors has just released a new generation of humanoid robot IRON. He Xiaopeng, chairman of Xpeng Motors, said that based on Xpeng's accumulation of L3 level autonomous driving and world models, Xpeng expects to be able to mass-produce L0 level humanoid robots for industry in 0 years.

In terms of cooperative research and development, BYD, Geely, Lynk & Co and other companies have chosen to cooperate with UBTECH, a robot company. It is understood that the industrial version of the Walker S series developed by UBTECH has entered a number of car factories for practical training. The technical person in charge of UBTECH revealed that the product has completed the first stage of stand-alone autonomous intelligent training. With the opening of the 0.0 stage of training, UBTECH will accelerate the iterative upgrading of products and technologies through data accumulation and model training of multi-machine collaborative training.

Ji Xuehong, director of the Automotive Industry Innovation Research Center of North China University of Technology, told reporters that humanoid robots and self-driving cars have commonalities in the underlying technology, and the layout of humanoid robots by car companies will help better play the technical advantages of enterprises.

"Some technology-based car companies regard humanoid robots as one of the strategic core businesses, and tend to control the whole industry chain independently, such enterprises need to develop their own key technologies and reconstruct the supply chain system, especially in the stage when the industrial ecology is not yet mature, and the ability to be independent is a competitive barrier. At the same time, some vehicle companies focus their resources on automotive product development and intelligent driving, and do not plan to invest in the research and development of humanoid robots on a large scale, and it is a more realistic choice to indirectly deploy through strategic cooperation or capital participation. Ji Xuehong said.

Cost reduction, efficiency increase and technology synergy two-wheel drive

Behind the large-scale layout of car companies, the industry generally believes that the participation of car companies in the humanoid robot track is expected to bring huge synergistic development effects.

On the one hand, automobile production, as an industrial scenario with many production links, high degree of standardization and large capital investment, provides a good training environment for humanoid robots, and is expected to accelerate the technology iteration of humanoid robots.

You Wei, chairman and general manager of EFORT Intelligent Robot Co., Ltd., told reporters that for humanoid robots, the current operation in industrial scenarios is relatively standardized and structured, and there are more professional users. "At present, we are working with car companies to build a test platform for typical applications, such as part of the final assembly of automobiles, to strengthen the training of humanoid robots. I believe that in the future, the improvement of the intelligent level of humanoid robots will need to go through the process of From B to To C, and the application of industrial scenarios such as automobile manufacturing is expected to achieve large-scale popularization faster. ”

At the same time, the accelerated implementation of humanoid robots also brings ideas and space for further cost reduction and efficiency increase in automobile manufacturing.

"From the perspective of the underlying technology of humanoid robots, including core technologies such as artificial intelligence, motion control, recognition and perception, car companies with self-developed capabilities are generally mastered. Humanoid robots are expected to solve the problems in the process of production line automation and achieve further cost reduction and efficiency increase. Zhu Jiangming pointed out.

Zhang Aimin, head of the robot R&D team of GAC Group, also said that the automobile production line is a typical scenario for the application of humanoid robots, and after a period of training and iteration, GAC Robot will give priority to entering its own automobile production line for testing, sorting, gluing and other work to solve the labor shortage of the production line and reduce labor intensity. "For GAC's existing automotive terminals across the country, relying on the advantages of GAC's automobile machine interconnection technology, GoMate can efficiently and smoothly complete daily work such as car charging, cleaning, and luggage handling. These capabilities will promote the transformation of automotive after-sales services from traditional labor-intensive to intelligent and automated. ”

"In the field of industrial manufacturing, represented by vehicle production, there are a large number of scenarios that require robots to improve production automation and intelligence, which provides a natural testing ground for humanoid robots." Mu Yao, assistant professor of the Institute of Artificial Intelligence of Shanghai Jiao Tong University, pointed out in an interview with reporters, "The versatility of humanoid robots with the blessing of embodied intelligence is expected to promote the upgrading of robots in factories from 'special aircraft' to 'one machine with multiple functions', which can not only improve production efficiency, but also promote the comprehensive industrial upgrading of industrial manufacturing fields such as automobile production, and realize more intelligent and flexible automatic production solutions." ”

Breaking through application bottlenecks requires multi-party collaboration

In the development of the humanoid robot industry, how to break through the bottlenecks of data collection and limited training scenarios, and further accelerate the application in the field of automobile manufacturing are also problems that all participants need to face.

"As the physical carrier of embodied intelligence, the hardware capability of the humanoid robot determines the lower limit of the service, and the intelligent algorithm determines the upper limit. However, at present, the further training of humanoid robots still faces problems such as the high cost of real-world data collection. Mu Yao mentioned, "The large model 'brain' of humanoid robots requires a large amount of reliable data for training, but humanoid robots lack efficient and low-cost data acquisition channels in industries such as unmanned driving." At present, in terms of real-life training, it is also necessary for car factories to build special real scenes, and professionals are required to control them, which greatly increases the cost of data collection. ”

In addition, the construction and cultivation of technological ecology is also a bottleneck for humanoid robots to be broken. "China has advantages in hardware manufacturing and cost control, but the software ecology is still a shortcoming, and a breakthrough is urgently needed." You Wei said.

To break through the current problems faced by the industry, experts point out that multiple parties need to work together.

"On the one hand, data collection can be simulated with more virtual simulation scenarios. At the same time, the real training scenario can be led by the government and work with multiple related enterprises to build a common production reality training scenario, so as to reduce the training cost and further accelerate the landing application of humanoid robots. Mu Yao suggested.

You Wei proposed that the government needs to take the lead in promoting the construction of the underlying application ecology, especially the joint industry partners to build a general technology base for intelligent robots. By continuously enriching the underlying skill library and algorithm library, improving the development tool chain, and greatly reducing the development threshold, experts in the application field can develop application systems without being proficient in robot control technology, so as to quickly establish a humanoid robot industry ecology.

Industry-university-research cooperation is also one of the keys to technological progress. Zhang Xiaohui, chief engineer of Topstar, proposed that universities, as technology research and development institutions, need to strengthen the research on embodied intelligence basic models, while enterprises need to further open up the control system of robot products, and build a localized technology ecology through in-depth cooperation with AI companies.

He Xiaopeng believes that we should learn from the early market cultivation and promotion experience of China's new energy automobile industry, and accelerate the cultivation of the humanoid robot market and the release of demand potential through the guidance and support of market promotion policies.

□Reporter Yuan Xiaokang, Wang Feng, Wu Tao

Source: Economic Information Daily