Embodied intelligence with nails.
Author|Zhao Jian
In February 2023, Chen Junbo, the former head of the Robotics Lab of Ali Damo Academy, left Alibaba to start a business and established a robotics company "Youlu Robot", positioning itself as a general brain of embodied intelligence.
Chen Junbo is joined by several veterans of the robotics and AI industry, including Gu Zulin, the former CEO of Alibaba Robotics.
There are deer robots standing on the tuyere of embodied intelligence. From the first day of its establishment, Youlu Robotics has taken product commercialization as the first task of research and development. Chen Junbo once said: "Embodied intelligence needs to obtain data in the real physical world like Tesla to improve the world model, but if you want to obtain the data of the real physical world, you must be the first to complete the commercialization and put the product on the market as soon as possible, so that you can take the lead in completing the world model with the language model." ”
Under this strategy, there are deer robots that have made rapid progress. At the end of May 2024, Youlu Robotics disclosed more than 100 million yuan in financing, with investors such as Innovation Works, Yuanjing Capital, Baidu Venture Capital, etc.; and an order of more than 10 million yuan for the embodied intelligent universal brain, which comes from a Chinese head cleaning equipment manufacturer. This is the first high-value order for a new startup in the hot track of embodied intelligence in the strict sense.
There is a deer cleaning robot AI130 at the Liangzhu site
Gu Zulin said: "To be honest, although this is the rhythm set when the company was founded, when we actually did it, it was still a little bit more than expected. Because, doing any project will be delayed by default, but there is no delay for the products of Youlu Robot. ”
There are many reasons behind this, and a very important reason is that compared with traditional manufacturing, Youlu Robotics is a digitally native enterprise. Among them, digital tools represented by DingTalk Docs, Yida low-code and AI Agent have become an endogenous part of Youlu Robot.
The digital native organization and management methods and business iteration methods have helped Youlu Robotics quickly run through the complete closed loop of R&D, product and commercialization in the first year of its business.
1. Open the company on the document
Chen Junbo developed the habit of writing documents at Alibaba.
On the first day of the establishment of Youlu Robot, Chen Junbo opened the company on the document. He created the first folder called "Youlu Robot" in DingTalk, and then set up three subfolders in it, called "Products, Technologies, and Algorithms", respectively @ three co-founders. After doing a good job in the division of labor, the three joint creators set up documents in their respective folders, and put a more detailed division of labor @ to each direct person in charge.
Layer by layer, clear and distinct. In Chen Junbo's top-down dismantling of business and division of labor, Youlu Robot opened the company in the document.
Here's a hiccup. Chen Junbo and the founding team are almost all from Alibaba, and they have been using DingTalk to write documents. Chen Junbo, who has been doing "half of his life" in Ali, uninstalled DingTalk as soon as he left the Ali system, and he wanted to experience the "freshness". But after using other document products for two months, Chen Junbo downloaded DingTalk back.
It's still easier to use DingTalk documents.
In Youlu Robotics, the company's business is fully documented. When you see a customer, you have to write a document, and you have to write a document for a meeting. Taking a meeting as an example, the team needs to build the document one day in advance, list the issues that need to be discussed, and start the meeting the next day directly into the topic, "no document, no meeting".
Gu Zulin told "Jiazi Lightyear": "Every valuable word said by the employees of Youlu Robot is in the document; If it's worthless, it's all in the chat log; The rest is nonsense, only feelings, no information. Documentation is our lifeblood. ”
Moreover, unlike large factories that may set restrictions on documents layer by layer due to the principle of confidentiality, the lower the level, the less permissions, because the company is still small, the documents are open to all employees. Anyone who enters the organization will be able to see all the documents that have been written since the first day of the company.
The purpose of openness is to collaborate. As a physical intelligence enterprise, Youlu Robotics involves many business departments, including hardware, software, and AI. The software team includes front-end development, back-end development, UI design, product managers, etc., the AI team includes model training, perception, planning, decision-making, data infrastructure, data cleaning, cloud operation and maintenance, etc., and the hardware involves a series of businesses such as R&D, production, and supply chain management in the traditional manufacturing industry.
Departmental walls naturally exist between different departments. For example, if algorithmic advancements can eliminate a camera, a series of changes will be involved at the hardware level, from form factor design, procurement, molds to factory production, and a series of relevant personnel will be involved. How to effectively coordinate between different departments is quite a challenge.
In Youlu Robot, each department and each relevant person in charge must operate on the document, specific to the time, place, quality results that need to be delivered, etc., and synchronize it to all employees. At this point, the document is not even a simple document, but takes on the role of a task kanban. Documents are the center of business flow, and they are the reconstruction of productivity and production relations brought about by digitalization.
Working on documents brings the digitization of business. At last week's DingTalk 7.6 product launch, Chen Junbo concluded: "Every cell in the company should be digitized to make the best preparation for training the company's embodied intelligence. We are not running the company, we are training the company and letting the company run itself. ”
2. From 0 to 1, low-code helped find PMF
Youlu Robot is positioned as an embodied intelligent brain and is a product-driven company. Therefore, finding PMF is the first priority of making products in the early stage. In the process of finding PMF, Youlu Robot has experienced many twists and turns.
Youlu originally wanted to put the embodied intelligent brain on the outdoor commercial large-scale sweeping robot, and the business model was to sell the brain's software system to traditional sweeping robot companies. After a door-to-door sale, the customer doesn't pay at all. Subsequently, Deer Robot tried to package the software and hardware together and put it on the customer's sweeper to do a direct demonstration. While customers are impressed, they are still skeptical about the value that AI brings.
Gu Zulin said: "Chinese enterprise customers are not willing to pay for software. The labor cost of our products is generally tens of thousands of yuan. The customer's sweeper may only sell for tens of thousands, and he can't accept it at a cost of tens of thousands of dollars to buy a software brain. As for AI, many traditional concepts in the manufacturing industry also make it impossible for them to digest such avant-garde concepts at once. ”
The feedback from customers forced Youlu Robot to make a change: from selling brain software, to directly making brain hardware, and selling finished sweepers equipped with brains, to "make a sample" for the market. Youlu Robot purchased several sweepers from the original customer in reverse, and directly "loaded" the self-developed brain into it and launched it to the market as a complete machine.
It was not an easy decision. Youlu Robot is a team with AI scientists as the core of software genes, if you want to make the whole product of the sweeper, you need to cut into the manufacturing from AI and software, and the organizational structure operation and management of the team up and down have entered an experience blank area. In the subsequent R&D, production and manufacturing management process, the team encountered various unexpected problems.
The first step is the R&D and testing phase. Youlu Robot itself is not from the cleaning industry, and will encounter various fragmented and divergent problems, either the height of the brush plate does not match, or the texture of the bristle procurement is problematic. It seems that there will be different reasons behind the same problem, such as the car lights are not lit, it may be the failure of the lights themselves, or it may be a software bug.
In addition to solving the technical problems themselves, it is more important to do a good job in process management, including problem alignment in multiple scenarios, information alignment between customer on-site testing and back-end production and research, and how to maintain the consistency of R&D versions. Essentially, it's a question of collaboration and the efficiency of information transfer.
In the mature manufacturing industry, this set of production and manufacturing processes will be solved by special ERP and MES systems. However, the manufacturing system of millions or even tens of millions of dollars at every turn is obviously not suitable for Youlu Robot, which only uses a dozen engineering application vehicles for R&D and testing.
Fortunately, Gu Zulin himself is from an IT background, and he quickly thought of DingTalk's low-code capabilities, which can quickly build applications through simple "drag-and-drop", which is very suitable for low-cost processing of fragmented and changeable production and manufacturing process problems.
In January 2024, after about a week of research, Gu Zulin built the first low-code application with DingTalk. After you are proficient, you can set up a process in half an hour. After it is built, it can be quickly verified, and if it does not meet the requirements, it will be modified immediately, and "where is wrong", it can be used in almost two iterations. The entire building process will not take more than one day, and the cycle of testing, iteration, and release will not exceed two weeks.
Gu Zulin himself built seven or eight low-code applications with DingTalk. He described low-code as like a small program, not the pursuit of big and complete but small and many, even if a low-code is only used once a year, it represents a process of the company.
Through the low-code, Gu Zulin has explored a set of online management processes exclusive to Youlu robots in the production, manufacturing, R&D and testing links: from the problem description, collection and sorting at the customer's site, to the solution given by the technical team, every step of the maintenance operation on the site will be recorded online, forming a complete file, and handed over to the product team for iterative upgrading. At the same time, the product team further transmits to the inventory management team through the work order system to ensure the timely deployment of maintenance materials and the rapid response on site; Or ask the relevant personnel to inspect other vehicles as well to ensure that all vehicles meet the same standards.
Gu Zulin told "Jiazi Lightyear": "In the manufacturing industry, it is completely different from 100 people engaged in production, and 10,000 people engaged in production. The process of ten people is the most painful, it is impossible to spend millions of dollars to get on a set of ERP, even if you are on the vendor, it is likely that you will not care about you. In the past, it was basically used to grit your teeth and get through, but DingTalk Yiqi low-code perfectly solves the needs of manufacturing enterprises at this stage. Moreover, DingTalk can be connected with many internal systems, such as R&D systems and data platforms, so that the internal flow of data is very smooth. ”
This kind of flexible process construction has made the Youlu robot team experience the feeling of "my life is up to me, and I am in charge of my own business".
Finally, after about a year of R&D and testing, Youlu Robotics released its first complete product, AI130, at the Shanghai CCE World Cleaning Expo in March 2024. As soon as the product was released, it became a hit in the sweeper market, creating an income of about 3 million yuan.
The news spread to the customer side, and the customers rekindled their interest in the products of Youlu Robot, and threw orders for embodied intelligent brains to Youlu Robot. The customer's attitude towards the embodied intelligent brain has changed 180 degrees.
3. Embodied intelligence also requires AI agent
In addition to the digitization of documents and processes, Youlu Robot is also exploring AI Agent, which is one of the important directions for the implementation of large models.
Gu Zulin said that all embodied intelligence enterprises will encounter two common problems, one is that AI Agent is originally intangible and is in the digital world; Embodied intelligence is a "shell" of the physical world, and it is the carrier for AI agents to work from the digital world into the physical world. Second, as long as software-related products are involved to serve commercial and industrial scenarios, there must be a terminal, which is difficult to bypass DingTalk, and the linkage with DingTalk is an inevitable trend.
Youlu Robot's definition of embodied intelligent robots is robots that think, interact and work like humans in the real physical world. If real humans need collaboration, then robots need collaboration as well. The collaboration of human beings with DingTalk, the management and collaboration between humans and robots, is DingTalk AI Agent + Youlu Robot.
At the product launch conference of DingTalk version 7.6, Youlu Robot and DingTalk jointly released the first embodied intelligent assistant "Youlu Robot" in DingTalk's AI assistant (i.e., AI Agent) market.
Gu Zulin called it the world's first cooperation case between embodied intelligence and AI Agent. Its cooperation model is not complicated: by DingTalk to open the API interface to enterprises, Youlu Robot accesses its own business system through DingTalk's interface, and then creates an AI assistant for online management of "robot employees".
One of its typical application scenarios is that when enterprise customers want to clean up outdoor garbage, they only need to give instructions from the "Youlu Robot" AI assistant on DingTalk, and the background will automatically dispatch the corresponding sweeping robot to clean it, making up for the lack of manpower of the cleaning aunt.
DingTalk itself has accumulated a large number of B-end enterprise users, for example, a considerable proportion of the property industry is also using DingTalk for collaboration, and these enterprise users will also be potential users of Youlu robots.
Therefore, DingTalk actually plays the role of a channel provider, and this channel provider is currently free. DingTalk said on the release that DingTalk is committed to AI inclusiveness, and the current AI assistant implements a commission-free policy and will not be charged from the cooperation between enterprise customers and Youlu Robot.
For this cooperation with AI Agent, Gu Zulin believes that Youlu robot represents a new quality of productivity, which is essentially to provide the supply of labor. DingTalk is a tool that greatly changes productivity, and if this tool can release more than 50% of productivity, the tool itself becomes productivity.
Gu Zulin used to be the CEO of robotics at Alibaba. He learned from SoftBank that Japan's strong development of robotics technology is due to the labor shortage. China is also facing the challenges of an aging population and a shortage of labor, so it is crucial for China to develop new quality productivity, especially technologies that can replace labor. This is also what the Arishi Robot is doing.
As a new quality productivity, DingTalk has two levels, one is the synergy itself, which changes the production relationship through digital means. Gu Zulin believes that the openness of documents is the only way to improve organizational efficiency, otherwise the concealment of information transmission will make things very complicated.
The second is DingTalk's AI assistant, which adds a large amount of digital productivity to the production relationship, which will be DingTalk's greatest competitiveness in the AI era.
Gu Zulin believes that with the blessing of AI assistants, the company's organizational structure will continue to change. The AI assistant we see today won't be its ultimate form. With the development of productive forces, production relations and organizational structures also need to be adjusted accordingly.
4. The next step in embodied intelligence
Today, Youlu Robotics has taken the first step towards embodied intelligence.
After the establishment of the company, Chen Junbo also wrote down the enterprise development route after repeated thinking and demonstration in the DingTalk document, and clarified the "WAH platform" of Youlu Robot.
W-Whale, called wheeled robots, is a very well-defined stock market, including forklifts, floor scrubbers, outdoor large lawn mowers, and so on. The intelligence of these products is a clear development direction, and many business scenarios have been verified by the market.
A - Composite type, the lower part of the product is the wheel, and the upper part is the mechanical arm. Youlu Robotics launched product research and development this year, and will officially release this type of product this year. One of them is a takeaway robot, which can automatically pick up packages and takeaways, run 20 kilometers per hour on the road, and automatically press the elevator and get on the elevator when it arrives in the community.
Gu Gulin calculated that there are about 13 million delivery workers in China. From the perspective of manufacturing, the stock market of 13 million represents the production of more than 1 million robots per year, which is the scale of an automobile group.
H——Human，就是大火的人形机器人。
Gu Zulin said that the commercialization of wheeled robots has begun, and the products in the next two stages are still in the research and development stage, and there are still big problems in algorithms and data.
Overall, both Youlu Robotics and the entire embodied intelligence industry are in a relatively early stage. Therefore, in the process of product iteration and commercialization, there will be many new and unexpected changes.
Youlu Robot has opened up a model of "Youlu Park Product Line" from the very beginning of the embodied intelligent brain.
The business model of the embodied intelligent brain is easy to understand, which is to sell the product to more customers to achieve scale. In this process, Youlu Robot participates in the production, manufacturing and sales of the whole machine as a supplier, and the brand of the whole machine is still the customer's brand, but it can also be open to customers under certain circumstances.
As mentioned earlier, in order to find PMF, Youlu Robot had to lead a complete product AI130 of Youlu Robot brand from 0 to 1, and achieved market success. This, in turn, has led some customers to propose a cooperation model to obtain the authorization of the Youlu Robot brand.
Gu Zulin said that in this way, Youlu Robot has not only done technical empowerment, but also brand empowerment. If the brand empowerment model runs through, Youlu Robotics itself does not need to invest heavily in the factory, but only needs to focus on technology iteration, product design and brand building.
This has the potential to lead to cooperation across the entire industry chain. Gu Zulin said: "For example, in a property cleaning scenario, the property company needs both park security patrol robots and sweeping robots. In the past, they were two independent brands, but after the emergence of Youlu Robot, if both brands are interested in obtaining Youlu brand authorization, their products have actually become one brand, and their sales channels can also be shared, and they can provide end-to-end services for property customers at the same time. In this way, ecological barriers are erected. ”
This model has actually been successful in the industry, and that is Xiaomi. Xiaomi empowers the entire IoT industry chain through its strong brand, bringing a win-win effect.
Youlu Robotics is exploring this model and has the opportunity to build a Xiaomi in the field of embodied intelligence.
In just one year since its establishment, Youlu Robotics has completed the PMF, the online release of the product and the initial commercialization, which is not fast.
Gu Zulin said: "To be honest, although this is the rhythm set when the company was founded, when we actually did it, it was still a little bit more than expected. Because, doing any project will be delayed by default, but there is no delay for the products of Youlu Robot. ”
One of the important reasons for this is that compared with the traditional manufacturing industry, Youlu Robot is a digitally native enterprise, and all businesses are iterated in a digital way, and DingTalk documents, low-code and AI Agent have become part of the endogenous business.
In a scenario that many people don't see, digital natives have brought profound changes to traditional industries.