From March 30th to 31st, 2024 China Body Intelligence Conference (CEAI 2024) was held in the west bank of Xuhui. The theme of this conference is "Living with the Body, Shaping the Future Intelligently". At the opening ceremony of the first day, the Shanghai Smart Industry and Innovation Alliance was officially launched. The initial members covered upstream and downstream enterprises in the industrial chain, as well as governments, universities and research institutions.
According to china computer federation’s definition, Embodied Artificial Intelligence involves artificial intelligence, robot learning, computer vision and other disciplines. The essence of embodied artificial intelligence is to emphasize that agents with physical bodies can get intelligent artificial intelligence research paradigm by interacting with the physical environment.
At present, as a frontier hotspot in the field of artificial intelligence, embodied intelligence is gradually moving from theory to practice and from laboratory to reality. However, many people are still unfamiliar with this huge potential "blue ocean" market. What can embodied intelligence do and change, and how is it related to the big model? Behind these questions, it indicates that software and hardware are coupled, symbiotic and mutualbestowcanOur embodied intelligence is rapidly entering human production and life.
Conference site
Why now?
At the 2023 World Artificial Intelligence Conference, Generative Artificial Intelligence (AIGC) became the absolute protagonist, and the attention of "embodied intelligence" also soared. But non-industry people are confused, "I heard it for the first time."
What kind of "intelligence" is embodied intelligence? Lu Cewu, a professor at Shanghai Jiaotong University, provided a definition. "In layman’s terms, it is the intelligence with the body, and the computer can be regarded as the representative of’ intelligence without the body’." In his view, it is timely to focus on physical intelligence. From a scientific point of view, the level of human understanding of itself, vision and motion control technology has been able to form a scientific category around embodied intelligence. From the industrial point of view, the software and hardware technologies involved in embodied intelligence have initially matured, and the whole industry is ready to go.
Dai Qionghai, academician of China Academy of Engineering and chairman of China Artificial Intelligence Society (CAAI), said that this year’s government work report pointed out that research and development applications such as big data and artificial intelligence will be deepened, and the "Artificial Intelligence Plus" campaign will be launched. Embodied intelligence is precisely the forward direction of artificial intelligence technology to the physical world. It is not only the integrated application of machine learning, natural language processing, computer vision and other technologies, but also the challenge and exploration of humanoid ability and humanoid wisdom.
The explosion point is now. Jiang Changjun, an academician of China Academy of Engineering, said that companies represented by OpenAI have built large-language models such as ChatGPT by using a large amount of Internet text and image data, which demonstrated the ability of AI to solve various tasks for the first time, and also showed human beings the potential of transcending domain restrictions and realizing general artificial intelligence (AGI).
"Although big models such as ChatGPT have demonstrated the interactive ability to deal with multimodal data and meet human needs, this interaction is still limited to the digital world." Jiang Changjun said that in order to realize truly universal artificial intelligence, it is necessary to give AI the ability to interact in the real physical world. This means that AI should have entities and have the ability to perceive, think and act like human beings.
Songling robot
As the combination of artificial intelligence and robot, embodied intelligence is regarded as an important part of new productivity. Last May, Huang Renxun, CEO of NVIDIA, a chip manufacturer, also said that the next wave of artificial intelligence would be embodied intelligence, that is, intelligent systems that can understand, reason and interact with the physical world.
Jiang Shuqiang, a researcher at the Institute of Computing Technology, Chinese Academy of Sciences, believes that the models trained by Internet data such as ChatGPT and Sora tend to be "off-body intelligence", and embodied intelligence needs to enter the real physical world, empower various industries from people’s livelihood to aerospace, and give feedback to the real environment, which means that there will be a place for embodied intelligence in thousands of industries.
Association with the big model
A week before the conference of embodied intelligence, the global developer pioneer conference was also held in the same place, the west bank of Xuhui. The discussion on the big model a week ago has not dissipated, and the "new explosion point" with intelligence has appeared one after another. What is the relationship between embodied intelligence and big model?
Lu Zewu said that the big model is more like a tool than a "brain" with physical intelligence. "Just as Newton’s law will be used to solve mechanical problems, embodied intelligence will have its own’ law of solution’, that is, its own big model." However, unlike the big language model and the big visual model, the training cost of real scenes is very high, and it is impossible to run at a high speed 24 hours as in the digital world. Therefore, the big model of embodied intelligence is bound to have a new training framework, which is the direction that Lu Cewu and his team at Shanghai Jiaotong University are studying.
Songling robot Shu Shu photo
At the conference that day, the guests mentioned that in the last round of generative artificial intelligence represented by ChatGPT and Sora, North America achieved a relatively obvious leading position. However, in the field of embodied intelligence, although China’s technology start and wave emergence are a little later than North America for about three years, at present, the world is basically at the same starting line of technology, all of which are before the outbreak.
In Lu Zewu’s view, China currently has two major advantages. One is robot hardware manufacturing, and the other is the number and richness of application scenarios. "The warehousing and logistics, food processing, catering industry and medical service industry have all started to use embodied intelligence. Perhaps after 20 years, people will have home-based pension robots to take care of themselves and their families. These are all embodied intelligence."
How to turn these advantages into "prosperous times" requires the joint efforts of all walks of life in Industry-University-Research.
The reporter noticed that in the exhibition area of the conference, many enterprises have moved out of the "magic weapon", including customer service robots that can communicate and talk smoothly, quadruped robots that adapt to complex terrain, and industrial robots that match various assembly lines. Robots equipped with AI big model "brain" can be engaged in domestic service, old-age care, education and medical care, facility inspection, emergency rescue and disaster relief, and show the latest application results of embodied intelligence in general scenes, professional fields, vertical industries and other tracks.
At the booth of Shanghai Fuliye Intelligent Technology Co., Ltd., a dark gray robot with a handsome metallic luster is waving its arms. This robot is called GR-1, which is the abbreviation of "General Robotics". According to the on-site staff, at present, the enterprise focuses on the robot body, which has the core technical advantages from the hardware structure to the motion control algorithm.
The "intelligence" of this intelligent product is mainly manifested in two aspects: sports ability and brain interaction ability. In sports, you can jump, run, wave, walk and dance, which can meet the needs of different customers in different scenes. "We open the interface, and customers only need to complete the upper-level design based on our platform, so that robots can complete the corresponding work, which has been applied in industrial production and education."
Fourier universal humanoid robot
In the more vertical field, coffee robots and ice cream robots have been active in many scenes such as hotels and social meals. Lian Wenjie, sales director of Shanghai Lebai Robot Co., Ltd., said that the seven-axis retail robot and six-axis cooperative robot launched by the company have core technical advantages in the field of grabbing items below 3 kg. In Marriott, Seasons and other hotel restaurants, Lebai’s noodle cooking robot has served for more than 8,800 days and penetrated into people’s daily lives.
How does Shanghai act?
Shanghai is the earliest pioneer in the field of physical intelligence in China. Last year’s "Shanghai Action Plan for Promoting High-quality Innovation and Development of Intelligent Robot Industry (2023-2025)" proposed that it would focus on tackling advanced technologies such as body intelligence. Luo Dajin, director of the Shanghai Municipal Science and Technology Commission, said at the meeting that Shanghai will accelerate the breakthrough of key technologies and underlying technologies such as body intelligence and humanoid robots, and accelerate its March to a globally influential science and technology innovation center.
As an important bearing area for Shanghai to build a science and technology innovation center with global influence, Xuhui District has recently launched a "combination boxing" to support the development of embodied intelligence, including issuing the policy of embodied intelligence industry and jointly building Shanghai embodied intelligence innovation center with the 21st Institute of China Electric Power Science and Technology. The output value of artificial intelligence industry in the whole region has exceeded 80 billion yuan.
The "Support Opinions of Xuhui District on Accelerating the Development of the Body Intelligent Industry (Trial)" issued in February this year is the first district-level support policy for the body intelligent industry in Shanghai. Yu Linwei, deputy head of Xuhui District, said that last year, Xuhui seized the opportunity of big model development, and the urban area jointly built the first "model speed space" in China and the only big model innovation ecological community in Shanghai. Within six months, the first phase of 20,000 square meters of carriers ushered in more than 60 enterprises and institutions.
Mode velocity space Shu Shu she
On this basis, Xuhui pursued the trend, followed the strategic opportunity of upgrading and leaping development of intelligent robots this year, and took the lead in launching the policy of supporting the intelligent industry with body in the whole city, focusing on key technology research, open collaborative innovation, high-quality incubator construction, expanding the application of "intelligent plus", and promoting industrial upstairs, etc., to support major original scientific and technological innovations in intelligent terminals such as humanoid robots and their application fields, and to serve the incubation and growth of hard-tech enterprises.
"Now we have to overcome the challenges in three dimensions, namely, to knock down the supply chain price, to raise the market share and to raise the revenue performance." The "three small goals" mentioned by Lian Wenjie represent the real thoughts of many small and medium-sized artificial intelligence enterprises: the sunshine and rain provided by the government are enriched, and enterprises are responsible for their efforts to thrive.
The reporter noticed that Xuhui’s opinions on supporting the intelligent industry have many bright spots in the strength of financial support and the dimensions of support fields, taking care of enterprises of different scales. For example, for projects that lead the development of embodied intelligence or make subversive breakthroughs, the maximum amount of support will be 10 million yuan, and the same amount of support is also applicable to embodied intelligent products that have achieved certain commercial results. Xuhui will also give enterprises a subsidy of up to 5 million yuan for technologies with industry-leading and demonstration effects that are applied to scene construction or digital and intelligent transformation projects that meet the standards of "industry going upstairs". Recognized high-quality projects of intellectual space can enjoy a maximum of 30 million yuan of municipal special subsidies.
Lu Cewu said that the development cycle of large-scale models and robot products related to embodied intelligence is long, which requires government departments to give more support and attention to technology research and development investment and development teams in the initial stage. "’‘everywhere is robot’ is a big investment, but as long as we can cross the threshold of the initial stage and start to scale up, the snowball will get bigger and bigger."