Schaffen und Fragen – Shao Tianlan von Mecademic Robotics: Bei Embodied AI gibt es kein „Heldenmythos“, nur „teuflische Details“.

What does an outstanding enterprise look like, and what qualities do successful individuals possess? On their journey of struggle, what "pitfalls" should they be aware of, and what is the most important change?

Creation·Inquiry poses questions to some outstanding Huachuang entrepreneurs, investors, and industry experts, and hopes to share their insights with you.

The protagonist of this issue is Shao Tianlan, the founder and CEO of Mech-Mind Robotics. Mech-Mind is one of the enterprises with the most implementation cases and the highest financing amount in the global embodied intelligent robot field. Its self-developed robot AI brain + 3D vision products have been implemented across industries and on a large scale in many fields such as automotive, logistics, and heavy industry, with more than 15,000 units implemented globally and maintaining the first market share for five consecutive years.

Mech-Mind has formed a standard product and general components of the robot's "eye-brain-hand" based on 3D cameras + self-developed AI algorithms + software platforms + dexterous hands, which are widely used in various typical scenarios of industrial manufacturing and logistics, and are gradually extending to service and household scenarios.

As early as the beginning of 2017, Huachuang Capital exclusively led the Pre-A round of financing for Mech-Mind.

⍺

The company introduction PPT originally displayed on the projector was switched to JD.com's purchase record, showing a blue shirt. This is exactly the shirt that Shao Tianlan, the founder and CEO of Mech-Mind Robotics, is wearing during this interview. Since 2012, whenever he runs out of shirts, he will find the same historical order and buy a few more identical ones. As long as the shirt is not taken off the shelves, Shao Tianlan will keep buying the same one.

This shopping order, on the one hand, seems to be Shao Tianlan's self-verification: from his appearance and figure to the story Mech-Mind tells externally, there hasn't been much change compared to when he first started the business eight years ago. "No false claims, no deviation," Shao Tianlan is glad that since the company was founded in 2016, it has always been on the right track.

On the other hand, Shao Tianlan believes that men's clothing and industrial products follow a similar logic. Although they initially have high entry barriers, complex decision-making processes, and long verification cycles, they emphasize first-mover advantages. Once users recognize them, they will keep making repeat purchases. This is the advantage of the ToB industry - rational decision-making and analysis. Once standardized, it is not easy to change.

In early 2017, Huachuang Capital exclusively led the Pre-A round of financing for Mech-Mind. Although the PPT presented to investors back then can still be used, the changes that have taken place in Mech-Mind are also obvious: in addition to continuously iterating the robot's 3D vision "eyes" and AI "brain," based on the accumulation of operational capabilities, they have also developed a five-fingered dexterous hand. This not only endows the robot with the abilities of precise perception, intelligent decision-making, and efficient execution but also enables Mech-Mind's products to be implemented across industries and on a large scale in many fields such as automotive, logistics, and heavy industry, making it the largest unicorn enterprise in the global "AI + robot" field.

As the founder, Shao Tianlan has often self-deprecated himself in the past few years, calling himself the company's "Number One Customer Service," "Chief Product Manager," and "Entrepreneurial Dog." Because the industries served by Mech-Mind seem to be a trillion-dollar market from a distance, but up close, they are ten thousand one-hundred-million-dollar markets. The biggest challenge here is not to meet the specific needs of specific customers but how to efficiently meet the various needs of thousands of users.

Therefore, behind the seemingly cool embodied intelligent robots, there are actually a large number of trivial details to face. It is precisely the refinement of these details that constitutes Mech-Mind's huge assets and barriers, allowing the company to enter the game in the process of moving towards the ultimate goal of embodied intelligence.

On weekdays, Shao Tianlan likes to read biographies of entrepreneurs to see how these real startup companies take action before a consensus is formed. At the same time, he also realizes that people like Steve Jobs and Elon Musk are exceptionally talented and difficult for ordinary people to imitate. Fortunately, in the field of embodied intelligence, it never depends on a very small number of extremely intelligent people, nor is there any exclusive secret. This is exactly what fascinates Shao Tianlan: "We don't have a genius story, only fighting monsters and leveling up one punch at a time."

Oral account: Shao Tianlan, founder and CEO of Mech-Mind Robotics

Written and compiled by: The editorial department of "Creation·Inquiry" of Huachuang Capital

Don't engage in differential competition

Not long ago, we participated in the 2025 WAIC World Artificial Intelligence Conference. Over the course of a few days, Mech-Mind's booth became one of the most popular ones at the conference. It was still crowded at 12:30 noon and remained full of spectators until the very end of the conference. Everyone was watching how the robots folded clothes, acted as "pickers," and "salesmen."

Don't think that folding clothes is a simple task; it's actually quite difficult to do well. Since clothes are typical flexible objects, the robot needs to be able to efficiently execute long-sequence flexible and complex tasks. Its two arms must cooperate with high precision to complete the "pick-fold-place" process. Because clothes are too soft, the robot also has to learn to adapt to the situation and know where to smooth out the wrinkles. This is much better than me. Every time I travel on business, the shirts I fold myself are always wrinkled.

The "Dual-Arm Robot Mass Object Classification" demonstration shows the robot's ability to autonomously sort a large number of random objects. We will place dozens of objects such as toys, snacks, daily chemicals, and fruits in front of the robot. Their materials, shapes, and sizes are all different. The robot can understand and recognize labels on its own and perform real-time classification according to natural language instructions issued by humans. For example, if I place a mango, an eggplant, and an umbrella in front of it, it can place the objects under the labels of fruits, vegetables, and daily necessities respectively according to my instructions. This generalization ability can now meet the high-speed sorting needs of various fields such as industry, logistics, e-commerce, and food for a large number of objects.

During the WAIC, the humanoid robot's shelf picking was also very popular. After the audience placed an order for a drink on-site, the robot would walk towards the shelf on its own, pick up the drink, and then hand it to the audience. It could also adjust its height according to the height of the shelf. The Standford humanoid robot DARWIN platform performing this task is equipped with Mech-Mind's embodied intelligence "eye-brain-hand."

Although we don't manufacture robot bodies ourselves, we can be compatible with dozens of brands and thousands of different robot models on the market. Behind this is Mech-Mind's self-developed technology stack for general robots: the Mech-Eye high-precision 3D camera, the Mech-GPT robot multi-modal large model, and the Mech-Hand bionic five-fingered dexterous hand.

This is our first full-scale demonstration of the full-stack AI capabilities of the embodied intelligence "eye-brain-hand." Skills such as folding clothes and sorting goods, which seem easy for humans, require a very solid AI technology foundation for robots. Only through continuous iteration of our technology can robots master these key abilities.

Since its establishment, Mech-Mind has been using intelligent technologies such as AI and 3D vision to endow robots with more advanced sensing, perception, planning, and other abilities, and using general products to solve common needs. The reason for equipping robots with "eyes" is that human eyes are the most common in comparison. For example, I'm not good at playing football, but my eyes can judge the situation on the field; I can't twirl a pen, but my eyes can judge the shape of the pen.

In the industrial field, various processes such as assembly, cutting, and welding, whether for a ship dozens of meters long or a component a few millimeters in size, have commonalities in terms of visual operation - identifying the type of the object, judging its state, accurately positioning it, and then guiding the robot to complete the corresponding actions.

Currently, the high-precision camera we produce can reach a maximum accuracy of 0.2 microns. What does this mean? If you split a hair into 400 parts, each part is approximately 0.2 microns. In addition, Mech-Mind's 3D vision "eyes" can adapt to various lighting environments, object materials, positions, and postures. Even if the object is reflective or transparent, it can still produce high-quality images.

(Point cloud image of office supplies generated by Mech-Mind's Mech-Eye high-precision 3D camera)

Therefore, Mech-Mind's robot "eye-brain-hand" is versatile in both manufacturing and logistics: it can perform many tasks such as loading, basket stacking, welding, and quality inspection in the welding workshop of an automobile factory, and at the same time, it can also handle transportation in a milk powder factory or an express logistics station. Our customers come from a wide range of industries, from the manufacturing of ships dozens or even hundreds of meters long to the production of mobile phone components a few millimeters in size.

Currently, more than a dozen product SKUs of Mech-Mind can cover most usage scenarios. As for artificial intelligence algorithms, it would be too costly to develop, train, and verify them specifically for each scenario. Therefore, the most crucial part is still to make the commonalities in the "brain" part.

In July 2024, we launched the Mech-GPT robot multi-modal large model, which was jointly developed with Academician Zhang Jianwei, the founding technical advisor of our company. The Mech-GPT multi-modal large model is like installing a smart brain in the robot. You only need to have a natural language conversation with it, without any complex programming or professional knowledge, and the robot can understand and execute human commands.

Mech-GPT can also be compatible with various robots and tools, not limited to a specific type of hardware. In any field such as industry, logistics, retail, agriculture, daily life, and scientific research, robots can complete a variety of complex tasks, which also greatly reduces the threshold for using robots.

Based on the AI "brain" + 3D vision, we have also developed the dexterous hand Mech-Hand. Its size is similar to that of a human palm, but compared with traditional dexterous hands, it is smaller in size, has a higher degree of freedom, and stronger motion control ability. From precision electronic devices to irregular objects, Mech-Hand can grasp them all, and it can flexibly adjust the number of fingers needed according to the size and shape of the object. Similarly, Mech-Hand is not limited to "special machines for special purposes" but can be applied across different environments and tasks.

(Mech-Mind's dexterous hand Mech-Hand)

We standardize and generalize the robot's "eye-brain-hand" to the fullest extent and then cover all industries to meet various needs. The advantage is that once it is developed, there will be a strong scale effect and Matthew effect, and the product's efficiency, globalization, and customer coverage will all be excellent.

Different from consumer products that pursue differentiation, the fields we serve face the most common needs and the most objective standard products. I have studied those industrial giants, and they almost all target the mainstream market.

So, since the start of our business, we have never deliberately engaged in so-called differential competition. We focus on mainstream products, mainstream industries, mainstream applications, and mainstream customers. Whether it's visual software, multi-modal large models, or dexterous hands, both the technology and the product form are very mainstream, and our customers come from major industries such as automotive, home appliances, logistics, e-commerce, engineering machinery, and steel.

Running a company doesn't necessarily mean engaging in differential competition. Competing head-on in the mainstream market is actually more likely to produce giants.

The devil is in the details

Don't laugh at me when I say this. When I was studying and doing robot research in Germany, I thought it would be amazing if someone invested 100 million yuan in me. I could make the product technology top-notch and let everyone in the world own a robot.

Today, Mech-Mind has completed multiple rounds of financing and is the largest unicorn enterprise in the global "AI + robot" field. However, I have become more and more in awe of and humble towards the industry.

When I entered the industry more than a decade ago, I found that this industry was different from what I had imagined. In 2012, I graduated from the School of Software at Tsinghua University and went to the Technical University of Munich in Germany to study robotics. After graduation, I worked at a well-known German robot enterprise and participated in the research and development of the most advanced industrial robots at that time. I thought I would be dealing with high-end things like algorithms, artificial intelligence, and automatic control every day, but I ended up doing everything from screwing screws, soldering circuit boards, to repairing circuits.

During those years in Germany, I was like a full-stack engineer, having to know a little about everything from artificial intelligence algorithms to soldering circuit boards. This experience also made me understand that the robot industry is never a high-end industry; it is actually full of a large number of details.

Since starting the business, I have visited more than 200 factories. Even today, I still visit the sites of many domestic and foreign customers because you can't just imagine what the customers need. Only by going to the factory and having in-depth discussions with the workers can you know their real needs.

For example, moving a box seems very simple, right? But there are still detailed requirements: when stacking the boxes, the labels should face outward; the boxes should be stacked in an interlaced way to improve stability; and if there is a problem, how should it be solved?

We have encountered all kinds of ridiculous situations in these aspects.

I remember in the first two years of starting the business, we once received a call from a customer asking why our product wasn't working. I immediately sent someone there in a hurry, only to find that a big spider was lying on it, blocking the camera.

You can't blame the customer for this, let alone the spider. Later, to solve similar problems, we had to make the camera have good waterproof and dustproof effects. The lens should be able to withstand cleaning with brushes, brooms, or even water, and it should also have a self-monitoring function for abnormal imaging, etc.

In addition, we have also encountered situations where we thought the product was malfunctioning, but it turned out to be a broken network cable, a dead wireless mouse battery,