What does an excellent company look like, and what are the characteristics of successful people? In their struggle, what are the pitfalls that need to be paid attention to, and what are the most important changes? Chuang·Q asks questions to some outstanding Huachuangists, investors, and experts in the industry, and hopes to share their ideas with you. The protagonist of this issue is Shao Tianlan, founder and CEO of Mecamander Robotics. Mecamander is one of the companies with the most cases, the most comprehensive technology, and the highest amount of financing in the field of AI+ industrial robots in the world. Its self-developed sensors and software serve the automotive, lithium battery, logistics, 3C and other industries, ranking first in market share for three consecutive years, and its business covers more than 50 countries and regions.
"There's no hard work, it's just head-to-head confrontation," Shao Tianlan, founder and CEO of Mecamander Robots, recalled in an interview. His favorite robot-related movie is "Pacific Rim," which tells the story of humans fighting against the invasion of monsters. , creating giant mecha warriors to fight. Although the plot of the movie is simple and even mindless, it is a close-up hand-to-hand fight between robots that are taller than skyscrapers and monsters. But Shao Tianlan feels that this is very similar to the robotics industry - facing a large number of engineering and business problems cannot be solved by a few smart people coming up with some brilliant ideas, but can only be solved bit by bit with a lot of hard work.
This is what Shao Tianlan has deeply felt after working in no less than three to four hundred factories since he founded Mecamander in 2016. As a computer major, he clearly feels the difference between the two fields: the Internet track is relatively flat, but Manufacturing needs are very fragmented - not only are there many processes, links, and scenarios, but they are also highly fragmented. Each company's needs are also different. From the beginning of the company's establishment, Shao Tianlan knew clearly: "The manufacturing industry is a trillion-dollar market from a distance, and a market of ten thousand to one hundred million from a short distance. The biggest challenge in serving the manufacturing industry is not to meet the specific needs of specific customers, but how to be efficient. to meet the various needs of thousands of users.”
Since its establishment, Mecamander has used intelligent technologies such as AI and 3D vision to give robots more advanced sensing, perception, planning and other intelligent capabilities, and use universal products to solve common needs. If we can efficiently integrate 100 of the 10,000 or even 100,000 100 million markets, that would be a 10 billion market. The large AI model is expected to make further leaps in robot intelligence, and the robot market may be 10 times or even 100 times larger than the existing one. This undoubtedly injects great imagination into the future of Mecamander.
In order to fill the gap in the demand for non-standard automation in these markets, Mecamand uses its core technical advantages of AI + 3D vision to make robots into general-purpose products or infrastructure platforms to reduce the need for non-standard hardware. After seven years of exploration, Mecamander has become one of the companies with the highest financing amount, the most comprehensive technical capabilities, the most implementation cases, and the widest application fields in the world's AI+ industrial robot field.
Nowadays, with the rapid development of technology, AI large models are changing with each passing day, and this wave is also surging in the industrial robot industry. When the future market becomes objectively standardized and there is no room for differentiation of products, the outcome of industry competition can only be "head-to-head". In this regard, Shao Tianlan is optimistic: "The large AI model makes the industrial market flatter. In a market that cannot be differentiated, a few giants will eventually be formed. So in the next three to five years, we will continue to polish our technical products." , unlocking more industries and taking intelligent robots to the next level."
The full text is shared below:
Q: Hua Chuang Capital
A: Shao Tianlan, founder and CEO of Mecamand Robotics
Q1: At the end of 2016, you came back from Germany and founded Mecamander Robotics. What was the opportunity to start your business at that time? What opportunities do you see?
Shao Tianlan: 2016 is known as the "first year of artificial intelligence." At that time, whether Alpha Go defeated Korean Go player Lee Sedol or the development of computer vision, everyone paid great attention to artificial intelligence. At the same time, the domestic entrepreneurial boom is also surging. When I was involved in it, I didn't predict what the subsequent development trend would be, but looking back now, it was indeed a good time to start a business.
In fact, the original intention of Mecamander has not changed since the first day it was founded - to make robots smarter. My study and work in Germany all revolved around industrial robots, so I clearly saw the bottlenecks of robots. At that time, our graduate team of seven or eight people was doing robot tasks, and it took a week of programming to make the robot do some simple things. But if it were a human, it might only take a sentence or two to understand the task. Therefore, it has always been our goal to give robots better perception and planning capabilities.
In April 2017, Huachuang invested in our Pre-A round. In fact, the company had only been established for half a year at that time. It had no customers and few fully formed products, so Huachuang really recognized it at a very early stage. Got us. Now more than six years later, our products have completed early exploration and become large-scale, global applications. The company's cumulative financing has exceeded 1.5 billion. But making robots smarter through artificial intelligence technology is still what we are trying to do.
Q2: You also mentioned that 2016 is the first year of AI+Robot, robot body, robot arm... among many tracks, why did you choose the industrial 3D vision segment?
Shao Tianlan: From an industry perspective, the entire robotics discipline setting in the past few decades has been more focused on mechanics and control, whether it is Tsinghua University, Shanghai Jiao Tong University, Harbin Institute of Technology, or Japanese universities. But now, a large number of computer-related departments at Tsinghua, Stanford, Berkeley, and Munich, where I studied, are researching robots. Therefore, our generation's robot industry is actually more "soft", that is, there are more algorithms, software, and artificial intelligence work. This can also be seen from professional journal articles. In terms of quantity and progress, it is also more and faster than in the past era where machinery and control were the main focus. This is a big technological trend that has similarities with the automotive industry. Of course, subjects such as mechanics and control are still very important foundations and cannot be neglected.
The reason why I chose this field is also related to my past experience. After graduating from the School of Software at Tsinghua University in 2012, I went to Germany to study robotics as a graduate student at the Technical University of Munich. I then worked for a well-known German robotics company and was deeply involved in the entire process of advanced collaborative robots from research and development to birth. I've been learning and working on things related to robotics, and the field itself is very interesting. Of course, I will only work in this field (laughs).
Q3: Compared with the Internet industry, what is the biggest challenge for entrepreneurship in the industrial field?
Shao Tianlan: In my opinion, robot service industry in the industrial field is actually a more reasonable discipline, that is, the customer's needs and required indicators must match the form and performance requirements of your product.
The Internet faces people, and people's needs are difficult to describe. It is more like an art. Many things are difficult to predict based on reason. For example, WeChat and Douyin must have merits technically, but users’ use of these apps is not entirely based on technical considerations.
But what is the success rate for industrial robots? What speed is achieved? How intelligent is it? How much applicability, reliability, etc. - these are more objective standards, and for people like us with a technical background, they are actually something that is better grasped.
But conversely, the difficulty lies in the high requirements. For example, we built our own high-standard camera factory to produce sensors, and put a lot of effort into improving the production linkage rate. We must not only explore cutting-edge artificial intelligence technology, but also Ensure consistent and reliable manufacturing. From the hardware's optical and electronic algorithms, artificial intelligence planning to sales delivery, the entire chain is very long and demanding.
In contrast, the Internet is like "one handsome person covering up all the ugly ones", with more emphasis on "long board". However, our kind of robot products require more practicality and rigor. While having strong boards, they should not have excessive shortcomings, so it will also More difficult.
Q4: Mecamander mainly produces industrial-grade 3D+AI products. Can you tell us what stage your current core product categories and application layout are at?
Shao Tianlan: Since our establishment, we have gone through four stages of development: technology accumulation, product implementation and application, product iteration and scale, and market globalization.
At the technical level, we continue to iterate in terms of camera imaging, optical design, AI vision algorithms, robot motion planning, and grasping planning. Products include industrial-grade high-precision 3D cameras, visual programming visual algorithm software and deep learning platform software, etc., mainly focusing on the field of machine vision, and the technology revolves around sensing, perception, and planning. The company's core devices have been fully self-developed and produced, and a total of more than 350 patents have been authorized and applied for.
We have entered the large-scale application stage since 2019. AI+3D vision solutions have been implemented on a large scale in many fields such as automobiles, logistics, heavy industry, etc. Typical applications include: disordered grabbing, loading and unloading, depalletizing, inspection, high-precision measurement, Robot gluing, spraying glue, etc.
Currently, Mecamander's business has covered more than 50 countries and regions, with more than 3,000 implementation cases worldwide.
Q5: Mecamander will start globalization in the second half of 2021. For a startup, what is your global market expansion strategy? Which countries or regions will you focus on?
Shao Tianlan: We currently have four subsidiaries overseas, located in Munich, Tokyo and Seoul. Overseas business already accounts for a large proportion. Now the company as a whole is growing more than double every year, and it is much faster overseas, growing several times every year.
What we are selling now is the fourth generation product, which has ranked first in market share for three consecutive years. The fifth generation is also under development, so it can be understood that Mecamander has something similar to the App Store, which contains our various products. You can use the application directly when you get it overseas. Of course, in response to the different situations of each country, we will also localize accordingly. We hope to bring many mature experiences from China with us, but we cannot copy them mechanically. After all, the needs of the downstream companies we serve will also be different in each country. Same.
As for the key overseas layout, it will still be placed in developed countries, because the needs of these countries and regions are very urgent, and our products are also very competitive locally.
Q6: Companies in many industries now choose to go overseas in order to escape "involution". For your field, is going overseas also to find more growth dividends?
Shao Tianlan: Let's imagine that if China is not so "volume", then companies will not go out? In fact, there are no companies in our industry that only focus on the Chinese market. The industrial core devices we make are general-purpose products. Siemens, Keyence, and Cognex will not produce so-called local products. They must be global. .
This is especially true from a historical perspective. There are no local companies for general device companies in all industrial fields, whether in China, Japan, the United States, or Germany. The business logic behind them is the global market. If a company wants to capture 10% of the global market, and no matter how big it is in the domestic market, it can only account for 20%. In the end, it will not be able to keep up with others in terms of products and scale. Therefore, going overseas is not because the domestic market is too big. The passive choice we made is that standardized products like ours are destined to face global competition. Since this is the end result, we must keep the beginning in mind and enter the game in time.
Q7: In addition to the world's leading companies such as Siemens, Keyence, and Cognex, there are also traditional robot manufacturers, manufacturing giants, and start-up companies that are constantly entering the market. In comparison, where does Mecamander's differentiated advantage come from?
Shao Tianlan: In my opinion, this industry is destined to have no way to "differentiate".
First, in fact, most people’s “differentiation” is false. "Differentiation" means different target customer groups and target needs, which often means targeting niche and segmented customer groups and needs. If it targets the same customer group and solves the same needs, but the performance, quality, service, appearance, price, brand, etc. are different, it is not called differentiation. For example, if you just put a few more peanuts in Kung Pao Chicken to change the taste, it wouldn't be called differentiation. But if I make a sugar-free Kung Pao Chicken for diabetics, this is the difference.
Second, companies do not necessarily need to differentiate. In fact, it is easier for giants to emerge in the mainstream market if they confront each other head-on. Because the total volume of the consumer goods market is very large, a very good company can be built by segmenting it into small segments. But giants in the industrial field are almost all in the mainstream market.
Therefore, our company does not deliberately engage in so-called differentiated competition, but focuses on mainstream products, mainstream industries, mainstream applications, and mainstream customers. Whether it is 3D cameras, vision software, or AI, the product forms are very mainstream, and the customers we serve also come from large industries such as automobiles, home appliances, logistics, e-commerce, construction machinery, and steel. Technically speaking, artificial intelligence, 3D perception, robot planning, including multi-modal large models are also very mainstream technologies. So what we make are standard products, and we are not looking for a very segmented small industry to compete.
There is almost nothing we do that is deliberately differentiated, including company management. If one thing is different from everyone else, then I think there is a high probability that it will be wrong.
As for the advantages of Mecamander, we first seize the opportunity. If we do it too early, mainstream technologies such as deep learning and sensing have not yet appeared, so we cannot apply it in products; if we do it too late, we will not be able to exist in this market. What's going on?
Third, we have always been very firm about our product form, productization concepts and business model. They have not changed since the company was founded. I think this is also what we have done better.
Therefore, leadership in all aspects of technology, products, customers, and capital allows us to form a positive cycle, which is the advantage of Mecamander.
Q8: If there is no so-called differentiation, are you looking for the optimal solution among existing needs?
Shao Tianlan: In the visual field where we are located, the needs of the entire market are very diverse, and we need to make standardized products to serve more customers. As a start-up company, it is unrealistic to challenge the most difficult things right from the start. What we meet is the most common needs of the manufacturing industry.
For many consumer products, such as women's clothing, the designs of different brands and consumers' aesthetic preferences are very different. Different people have different opinions, so there is no optimal solution to this matter; but in the industrial field, such as Schneider and Siemens, they pursue objectivity. Standard products will eventually reach the optimal solution. If you deviate from the optimal solution, the market will punish you.
Therefore, in a market that cannot be differentiated, the indicators we pursue are very objective. It has a theoretical optimal solution, and it depends on who can get closer. And this is destined that the final market structure will be very unified, facing global competition, which is also the cruelty of our field.
Q9: According to your judgment, what will be the future competitive landscape of the entire industry? What is the future development trend?
Shao Tianlan: This thing is very simple. All things that cannot be differentiated and have significant scale effects will eventually form a small number of giants.
What we do is obviously a highly standardized thing, without any room for differentiation. At the same time, there is a very obvious scale effect. The subsequent competitive landscape must be that there will only be a few companies in the world. Of course, we are very optimistic about the entire industry as a whole, because now artificial intelligenceEnergy technology and robotics technology are still in a very rapid development period. When Huachuang invested in us, we said we were going to do this, but when can we scale it up? When will we be able to achieve the scale we have today? What was said at that time was considered ideal by everyone, but now it has been achieved. In the future, we will still have such room for growth, because the entire intelligent technology and robotics are still in an early stage of rapid development.
Q10: In a previous conversation with Xiong Weiming, the founding partner of Hua Chuang Capital, you mentioned that the emergence of large models has the opportunity to push robot technology to another level, and the market may be 10 times or even 100 times larger than the current one. Big, can you tell us what impact this has had on you and the entire industry?
Shao Tianlan: Large models are really very important and can bring a lot of changes, which can bring the intelligence of the robot to a higher level. Like what our robots have done before, many of them are single-point capabilities, such as sensing (can scan objects at high speed like eyes), perception (can identify the type, posture, and relationship of objects). Large models can bring higher levels Intelligence, such as task understanding and overall decision-making.
It is able to have more common sense and understand more complex tasks. Let me give you an example. For example, we have to eat at noon, which involves cutting vegetables, cooking, washing pots and washing dishes. This requires the robot to be able to identify cucumbers and potatoes and pick them up. How to combine these actions in the process? , requiring robots to have common sense and complex planning and reasoning capabilities, which previous single-point, functional artificial intelligence did not have.
So the large model excites us very much. I think it will open up a series of applications and scenarios just like previous advances in artificial intelligence, taking the intelligence of robots to a new level.
Q11: Specific to Mecamander, what new explorations have you made in terms of large models?
Shao Tianlan: Not long ago, we reached a cooperation agreement with the laboratory of Academician Zhang Jianwei of the University of Hamburg and are jointly developing a large robot model that comprehensively integrates vision, speech and language capabilities. This model will enable robots to sense and understand multiple signals in the environment and communicate with humans through natural language. The research and development results can greatly improve the intelligence level of robots, allowing them to better naturally cooperate and interact with humans.
In the era of big models, embodied intelligence is now a major focus of attention. Academician Zhang first proposed the concept of cross-modal learning robots, which achieve universal embodied intelligence by integrating a large amount of noisy and multi-source heterogeneous multi-modal perception information such as vision, hearing, and somatosensory.
Our cooperation is still in the scientific research stage, but in the future we hope to bring this large robot model into more complex scenarios such as restaurants, hotels, hospitals and industries.
Q12: There has been a lot of discussion about "embodied intelligence" recently, and there are various innovations in the industry, from RT1 and PaLM-E to Li Feifei's "VoxPoser" and the Deepmind team's RT2. There are new ones every once in a while. Now that the results have been released, what stage do you think the development of “embodied intelligence” has reached now? How far is it from true "embodied intelligence", and what key issues remain unresolved?
Shao Tianlan: In the past, "embodied intelligence" was still a kind of science fiction, as far away as the moon, but now it is like Mount Everest. There is an essential difference between the two. Although it is still very high, it has fallen to the ground. Just like climbing Mount Everest, people can climb up slowly and already have the technical routes to climb up.
Now that we have a large model, it is equivalent to the stage of building a ladder. The entire technical route has a general direction and we know how to get up. Just like the Wright brothers invented the airplane, the airplane they made at first only flew a few dozen meters, but only a few decades later, the Boeing 747 was born. Therefore, once a potential technology has made great progress, although there is still a lot of specific engineering work to be done and the workload will be very large, at least we have a relatively clear development direction.
Q13: Scenarios in the industrial field are generally more complex, so customization has always been serious. After the large model is released, can these customizations be eliminated? For example, if you give a robot an instruction, can it automatically complete all operations?
Shao Tianlan: Universal things are actually a very important prerequisite for true large-scale applications. We have seen it in history. For example, computers were originally used to calculate rocket trajectories, ballistics, and process population data. The real large-scale popularization is actually After the advent of general-purpose personal computers. The universality of computers has also greatly optimized its overall scale and cost, so that it can be truly promoted. There are many similar things, and many special-purpose devices must be formed into a universal product before they can really start to become popular.
For us, we must reduce the customization of software and hardware development. Of course, some configurations still need to be made. Just like computers are standard products, users will still install different software. We try our best to use more general capabilities, especially sensors and algorithms, to reduce non-standard customization, so that the cost is better, the cycle is shorter, and it is more adaptable.
Q14: Can it be understood that with large models and better algorithms, customization and SKUs can be reduced?
Shao Tianlan: It can reduce the workload required for customized design of the entire hardware in each scenario. Let me give you an example. For example, in an automation scenario, do you use a universal robot, that is, "standardized products + universal vision" to solve the problem? Or design non-standard, customized structures?
If the devices used are more standard, and the more sensors and algorithms are used in them, there will be more general-purpose robotic devices, so the customization cycle required and the risk of changing the entire application in the future will be lower, so automation should be as much as possible Using more standard equipment, such as machine arms, robots, and AGVs, it is a standardized product compared to traditional conveyor lines and customized structures. Therefore, whenever possible, use combinations and configurations of standardized products rather than many complex custom designs.
Q15: It sounds a bit like Lego. We also learned that there are many scenes in the manufacturing industry that have not been replaced by robots, and the penetration rate of robots per 10,000 workers in China is not high compared with many countries. What do you think is the reason? What aspects still need to be improved?
Shao Tianlan: The main reason is the lack of skills. Every time we face a demand, the overall cycle of debugging and delivery will be very long. Whether it is automation or robots today, the overall process is still very slow. The outside world often imagines that people need to be replaced in this link, and the robot can be used just as soon as it is brought over. In fact, this is not the case. It is very normal for an automation project to take half a year.
There are not so many new things in this world. Let me use an analogy. Today, the penetration rate of scanning QR code ordering in restaurants is about 80%. However, automated ordering has been done for a long time. At first, professional equipment was used to order, but it It has never really been popularized because this system requires a professional company to do it, which takes a long time, high cost, and is inconvenient to maintain. It was not until the advent of smartphones and mobile payments that restaurant owners could do it themselves using QR codes that the technology became available and accessible.
Another example is the washing machine. To a certain extent, the washing machine is also a robot, but it still requires professionals to install the pipes. How can robots become highly intelligent in the future? After we give it instructions, it needs to use human-like eyes to identify the scene and understand and perform tasks. Only when it reaches a certain standard and height, everyone will have a robot in the future, and the number of robots will even exceed that of humans, just like today's mobile phones. Same as computers.
Q16: The Mecamander team currently has more than 700 employees. What is the biggest challenge you face in management? What are your expectations for the company in the next three to five years?
Shao Tianlan: This is indeed a very big challenge for us, because there are different teams within the company and the cultures will be different. For example, the R&D team in Beijing is composed of masters and PhDs from the world's top universities and outstanding engineers from large companies. The overall atmosphere will be very relaxed, and there are even cats in the office. The production team has a very rigorous style, and the sales team in Shanghai, The only requirement for them is to satisfy their customers. Therefore, different teams must have different organizational cultures and management styles, and finally must unite to achieve the company's common goals. This is indeed a big challenge.
In the next three to five years, I hope that the company can maintain rapid growth. Multi-modal large models like the one we are doing now, including various advanced sensors, can have more applications, enter more industries, and take intelligent robots to a higher level.
