Recently, the "2024 Zhiyou Growth Camp Spring Class" officially kicked off at KLCII Building. The event was hosted by Zhongguancun Zhiyou Research Institute and supported by Zhongguancun Street. The event especially focused on the cutting-edge track of "embodied intelligence", aiming to create a subdivided exchange platform for entrepreneurs to collide ideas, exchange experiences, and activate value, and inject new vitality into the vigorous development of "embodied intelligence" in mainland China.
The creators of 23 outstanding projects from the embodied intelligence subdivision track, 6 heavyweight mentors and a number of previous alumni gathered at the scene. Everyone started a collision of ideas around the development trend of embodied intelligence, potential landing scenarios, entrepreneurial opportunities and the first step of entrepreneurship. In addition, during the event, there were two simulation sessions, "Dream Auction" and "Xiyun Roadshow", leading changemakers to immerse themselves in the key elements of the entrepreneurial process.
Group photo of the 2024 Zhiyou Growth Camp Spring Class
The 2024 Wisdom Friend Growth Camp Spring Class will issue a letter of appointment to the mentor
Wang Tianmiao, honorary director of Beihang Robotics Research Institute and president of Zhongguancun Zhiyou Research Institute, believes that "only love can promote us to create unique results." I've often thought that when a person is young, if they can truly love and show what makes them unique, that's actually a valuable asset. In this entrepreneurial voyage of embodied intelligence, in addition to love, we also need to deeply understand this era, clearly understand our own advantages, and strengthen our will, so as to move forward better. ”
Wang Tianmiao, Honorary Director of Beihang Robotics Research Institute, President of Zhongguancun Zhiyou Research Institute
Embodied intelligence in prophecy VS entrepreneurial choices in reality
Two months ago, Figure and Open AI in Silicon Valley caused an "industry earthquake"; Two weeks ago, Boston Dynamics on the other side of the ocean announced that it would say goodbye to the hydraulic drive mode and the advent of electric power. The plot about embodied intelligence is developing at a speed beyond people's imagination.
In the mentor roundtable session of the opening ceremony, Ying Fei, vice president of Zhongguancun Zhiyou Research Institute, served as the roundtable host, and Wang Tianmiao, honorary director of Beihang Robotics Research Institute and president of Zhongguancun Zhiyou Research Institute, Huang Tiejun, chairman of Beijing Zhiyuan Artificial Intelligence Research Institute and professor of the School of Information Science and Technology of Peking University, many vice chairmen of Xiaomi Technical Committee and general manager of Xiaomi Robot Division, Xu Weijia, Meituan logistics planning expert, Wang He, founder &CTO of Galaxy General, assistant professor and doctoral supervisor of Peking University, The five mentors had an in-depth discussion on the theme of "Embodied Intelligence in Prophecy VS Entrepreneurial Choices in Reality".
In the mentor roundtable session, from left to right, Ying Fei, vice president of Zhongguancun Zhiyou Research Institute, Wang Tianmiao, honorary director of Beihang Robotics Research Institute and president of Zhongguancun Zhiyou Research Institute, Huang Tiejun, chairman of Beijing Zhiyuan Artificial Intelligence Research Institute and professor of the School of Information Science and Technology of Peking University, many vice chairmen of Xiaomi Technical Committee and general manager of Xiaomi Robot Division, Xu Weijia, Meituan logistics planning expert, and Wang He, founder &CTO of Galaxy General Motors, assistant professor and doctoral supervisor of Peking University
- Prediction of the "landing" and "technology" of embodied intelligence
Many: We all know that in industrial applications, the demands on speed, quality and precision are very high. I summarized that there are three main precision requirements: movement accuracy, grasping accuracy, and semantic accuracy. If we want to apply embodied intelligence technology in the service or industrial field, I think all three of these precision needs to be above medium high accuracy.
However, there are still some technical challenges. For example, it is a challenge to achieve a movement accuracy of 1-3 cm, and it is expected that it will take about three years to break through it; It may take about three years to reach the level of one millimeter of grabbing accuracy; As for semantic accuracy, large models have been well solved in this regard, but the hardware accuracy related to it is not ideal at the moment. In summary, I think it may take 3~5 years for embodied intelligence to achieve mass application.
Wang Tianmiao: I would like to talk about the landing scenario, I think the scenario of embodied intelligence is expected to innovate and land is the "generalized intelligent workstation for industrial scenarios", which aims to realize the generalized application of technology through a product, and solve the generalization problem from standardization to semi-structure, and then to master the skills of "ordinary technicians"; Next, there is the implementation of business scenarios, because it involves semi-structured and fully structured data processing, the breakthrough in this field is not obvious at present, and I think there is huge potential in fields such as healthcare, logistics, and cleaning.
Talking about technological development, I think that in the future, embodied intelligence needs to be supported by multiple professional small models, not just a single large model, because we need multiple models to respond to different scenarios and needs; Secondly, due to the limitations of the speed, size, and cost of embodied intelligent robots, we also need to reconsider the integrated design of vision, touch, and joints.
Huang Tiejun: I think 2045 may be a critical time for embodied intelligence to surpass people in an all-round way. The "comprehensive transcendence" here is not a simple progress, but refers to the fact that the robot's capabilities will reach an unimaginable height. Over the next 20 years, this sector will be influenced by a confluence of factors.
First of all, there is the "computer" part, and the evolution of the brain is the key. Today's large-scale model technology is developing rapidly, which will greatly accelerate the speed of robots' common sense cognition of the world for embodied intelligence, which means that in the near future, no matter what industry robots are engaged in, their understanding and cognition of the world will surpass that of humans. Secondly, it is about the "electric eye" part, the progress of the visual system is also very important, now we have achieved the recognition and positioning of both eyes at a speed of 5 milliseconds, and in the future, with the continuous optimization of the system and the progress of the algorithm, the response speed will be faster, greatly exceeding that of humans. Finally, the humanoid robot proposed by Wang He in the previous paragraph is best abbreviated as "electric human", which I think is very concise, that is, an embodied agent that combines advanced brains, excellent vision, and machine bodies will have the ability to surpass humans. Although I can't make precise predictions, I believe that in the near future, the "electric man" will play an important role in several scenarios and will exceed most of our expectations.
Xu Weijia: From the perspective of the development process of robot applications from automation to intelligent warehousing and logistics, in fact, the automation industry has been developing for decades, starting from Amazon's latent KIVA robot in 12 years, and it took a very short time to break out of large-scale application in China, and the scale has been relatively large in a few years. So from the perspective of the development process, the development of humanoid robots and the development of soft brains will be relatively short, but if the hardware wants to meet the requirements of precision and efficiency, from the perspective of our robot application process, it will take a relatively long process.
Wang He: One of the practical questions facing the industry today is whether there is a way to produce cross-industry data in large quantities and that data can be processed by a general-purpose large model at once. One of the views held by Galaxy General is that simulated synthetic data has such a possibility. The generalized grasping skills we have trained have achieved a 95% success rate for stacked objects of any material and shape, and the data is completely dependent on synthetic data.
Giants like Nvidia are also using physics simulation to learn from multiple tasks. Therefore, we do not rule out synthetic data and simulators as a low-cost data solution that can be applied across scenarios. For small businesses, this is both an opportunity to learn robotics skills and a disruptive challenge like multi-skill embodied large models.
- When to start a business and what defines "good product"
Wang Tianmiao: My observation is that there are still challenges of openness in the current application scenarios from industry, business to home. Even if the current valuation of some embodied intelligence startups has reached 2 billion, 5 billion, or even 10 billion, this situation reflects more the optimistic expectations of the capital market for the future landing of products, which may have bubbles.
But real value creation is just beginning, especially when the technology can be implemented and solve real problems, especially when the iterative cycle of hardware has not kept up with the evolution of software or the brain, and we see the opportunity for embodied intelligence. So, it's not too late to start a business.
Many: It's certainly not too late to enter the market, there are no robots that can solve problems, and there are no embodied models that can actually solve problems. There are two ways we choose to solve the problem, time for space, and space for time. If you do the brain, it is essentially time for space. If you have enough money to do it, it is recommended not to make short-term tool-based applications. Tools will soon be eliminated, or do something that has a certain degree of versatility, and must have versatility.
Huang Tiejun: I coined a word called "smart device", which means that there are intelligent devices and intelligent appliances. I believe that in five years' time, there will be many intelligent tools that we can't even imagine today. If anyone can creatively create a new "smart weapon", which is more dexterous, ingenious, and cost-effective than the machines seen today, there will be great market opportunities. This will test everyone's creativity and innovative imagination, and I think the opportunity is huge.
For embodied intelligence, the progress of the large model's common sense cognition of the world may exceed many of our expectations, which means that no matter what industry the robot does, its understanding and cognition of the world surpasses that of people, which should be within a few years. After the embodied intelligence combines the brain, eyes, and body, it is stronger than human ability, and will exceed the expectations of many people, and will surpass human beings in 20 years at the longest.
Xu Weijia: From the perspective of Meituan's industry, we focus on services that revolve around people's food, drink, housing and transportation. From informatization to the Internet of Things to warehousing and logistics, a large number of machines and equipment have been used to solve some problems, and now they cannot be solved, such as receiving, shelfing, and picking operations in the warehouse, which are difficult to complete with machines.
The market for embodied intelligent robots is huge, but it is not necessarily what kind of robot it is, such as unmanned delivery of drones or humanoids, even if it is a wheeled quadrupedal bipedal, no matter what form, but to solve some of the things that we can't do with humans and machines now, this is the product we need.
Wang He: We talk about embodied intelligence products, my narrow definition here is data-driven, if it is data-driven embodied intelligence entrepreneurship, we must think about a few questions. The first problem is that the core advantage of data-driven is generalization, and its disadvantage is precision, so we should look for tasks that require urgent generalization capabilities, but cannot be too high for accuracy requirements for productization. The second problem is to have data, if the data comes entirely from teleoperation, then there will be a problem of quick but not generalized, and one of the moats of Tesla's FSD is its millions of cars and hundreds of millions of hours of real-time driving data. If there is no data source, or the cost of data collection is high, then the future commercialization of the product cannot be calculated; In addition to this, a good product also needs to measure the cost of hardware.
- About how domestic players can achieve "corner overtaking"
Wang Tianmiao: China's manufacturing supply chain system is relatively complete, so China does have this opportunity. In the future, the wider the range of scenarios, the greater the opportunities for the industry. To achieve this, innovation becomes the most critical element, and this includes innovation in technology and product applications. When these innovations are combined with China's strong supply chain strengths, they will inevitably lead to significant growth in the scale of the industry.
Huang Tiejun: For the future development, I am very confident, our manufacturing industry will be fine, which is our special advantage compared with other countries. No matter what form of robot is needed, what kind of device parts, we can roll it out, and roll out something that is very cost-effective.
Many: In fact, we have a product development methodology, the first step is to imitate and make a product that can be used, the second step is to exhaust its physical boundaries and roll the product to the optimal physical boundary, the third step is to solve the problem of continuity of enablement, and the fourth step is to solve the problem of application.
In the current market environment, there is indeed a huge opportunity to start a business in the embodied intelligence industry. But the key is that we have to be clear about the continuity of enablement issues that the product needs to address. In the short term, we need to solve the problem of data closed-loop, model training, and model management for 50 to 100 sequence-long complex tasks.
Xu Weijia: We have no doubt that the market is huge, and at the same time, we are also confident that China's manufacturing industry can reduce the manufacturing cost of the robot body to an ROI level that we can accept. As long as we meet the requirements of consumers, its mass production is a very rapid development process, so we are still very much looking forward to the development of embodied intelligence.
Wang He: I think embodied intelligence must be the world of us Chinese in the long run, because in the end we can mass production, we can make it cheap, we can have such a large market, China's current aging population, the future labor shortage must rely on this is robots to fill. Now it is very important to be an embodied intelligence start-up, and we must do really valuable things in a down-to-earth manner, and don't be led by some bad atmosphere to do some demos, do some things that are not generalized and similar to recording.
Entrepreneurship Starts: How to Break Through the "First Kilometer" of Entrepreneurship
In the "Alumni Sharing" session, outstanding alumni such as Liu Wuyue, founder and CEO of Bluepoint Touch, and Li Chaoyi, head of accelerated evolution and commercialization, shared their practical experience on how to break through the "first kilometer" of entrepreneurship.
"As an entrepreneur, if I am given another chance to start over, there are two aspects that I will focus on." Liu Wuyue said, "In the early stage of entrepreneurship, one of the key points is talent recruitment, especially when there is a lack of experienced personnel specialists, the CEO must personally build a good talent echelon; Secondly, it is the in-depth understanding of the industry, we encountered a lot of challenges in terms of resources and direction at the beginning, largely because of the lack of sufficient understanding of the industry, and when we established a solid relationship with the scene side, this was also a key turning point for our business. ”
Liu Wuyue, founder and CEO of Blue Dot Touch
Li Chaoyi, who is on the road from 0 to 1 entrepreneurship, believes that "when we have a good entrepreneurial idea and get financial support, the real challenge is to turn the idea into a product." ”
Unlike software product development in the past, the development of hardware products involves a complex and lengthy supply chain building process. Entrepreneurs need to establish cooperative relationships one by one, which will be a process of teamwork and continuous breakthrough; At the same time, as a start-up company, it is also necessary to focus on a core direction, and clarify which direction is reasonable and can generate value based on the advantages of the team, so as to promote the process of productization and commercialization. In short, breaking through the "first kilometer" of entrepreneurship requires recognizing its own advantages, focusing on it, building barriers, and accelerating.
Chaoyi Li is the head of accelerated evolution commercialization
Dream Voyage: Grasp the key elements of embodied intelligent entrepreneurship
In the two virtual sessions of the passionate "Dream Auction" and the "Xiyun Roadshow", the entrepreneurs deeply appreciated the wisdom and charm of how to skillfully combine key resources, quickly assemble the team, accurately define product power and create a unique business model.
Through 22 virtual asset auctions involving core technologies, elite teams and innovative business models, they intuitively felt the close connection between assets and value. At the same time, the rapid establishment of four embodied intelligent virtual entrepreneurial teams not only provides entrepreneurs with the opportunity to communicate with like-minded partners, but also invisibly deepens their understanding and practice of entrepreneurship.
The scene of the "Dream Auction" and "Xiyun Roadshow".
Empowering entrepreneurship and sending a message to the future. Zhang Ruijun, professor of Renmin University Business School and founding partner of Yarui Capital, summarized the growth camp. In her view, "The journey of an entrepreneur is bound to be full of ups and downs." Sometimes, entrepreneurs are bathed in the sun, but more often than not, they face the challenges of storms. On this road, some people may accidentally fall into the ravine, but it is the love of this career that brings everyone together, learns from each other and supports each other in the growth camp. I firmly believe that with this love, we will be able to bravely climb mountains, cross oceans, and finally, wait for the flowers to bloom with peace of mind. ”
Zhang Ruijun is a professor at Renmin University Business School and a founding partner of Yarui Capital
The holding of the 2024 Smart Friend Growth Camp Spring Class not only provides a platform for entrepreneurs to showcase, communicate and learn, but also injects new vitality into the hard technology innovation and industrial development of embodied intelligence. We look forward to working with entrepreneurs to meet challenges, explore the unknown, and create more possibilities in the future!
About Zhongguancun Zhiyou Research Institute, it is a private non-enterprise legal person unit led by Professor Wang Tianmiao, Zhongguancun Innovation Partner and Honorary Director of Beihang Robotics Research Institute.
As a high-tech think tank and an early-stage hard technology venture capital service platform, Zhongguancun Zhiyou takes scientists as the core to carry out a new model of "achievement transformation + technology incubation + investment acceleration". The platform deeply focuses on the innovative technologies and achievements of robotics and intelligent manufacturing, and is committed to discovering scientific and technological innovation stars, cultivating influential specialized, special and new small giant enterprises, and providing strong support for accelerating the development of "new quality productivity" and hard technology "stuck neck" problems!
Up to now, Zhongguancun Zhiyou has incubated and invested in a number of international and domestic leading technology enterprises. Among them, there are 3 listed companies, namely "the world's leading enterprise of intelligent short transportation and service robots - No. 9", "China's first stock of surgical robots - TINAVI" and "the first echelon enterprise of industrial robots - EFORT".