As a student of Academician Xu Xūróu, the founder of Chinese optics, Li Yuan has been doing one thing for eleven years: creating a pair of “eyes” for machines.
These “eyes” go up into the sky, down into the water, and onto the roads, and are installed in drones, trains, cars, and various robots.
Today, as everyone starts discussing physics and AI, he is also rethinking: can these eyes help machines understand the world, in addition to helping them see?
From undergraduate to postdoctoral research, Li Yuan originally wanted to become a scholar. In 2015, he chose to return to his country to start a business and founded Benewake. A person who most wanted to be a scholar became CEO for 11 years.
In the first half of this year, Beixing delivered 250,000 laser radars for robots. The robot business revenue accounted for over half of the company’s total revenue, and this business revenue will double compared to 2024 in 2025.
Beyond these numbers, he was also thinking about what lidar could bring to machines.
"Eyes are not just tools for the brain, but also a source of knowledge." He said.
In his view, information such as distances and positions measured by lidar should, in addition to helping robots avoid obstacles, also serve as material for model learning. How these measurements can improve the model’s understanding of the physical world is the direction he is researching.
Li Yuan deliberately picks out something he doesn’t like to do every year.
This sounds like you are putting yourself in a difficult situation.
Li Yuan’s explanation is that many so-called “dislike” feelings stem from fear, or a lack of understanding. When something is too far away, one can only see a dark, long shadow.
Once you really get involved, things often become interesting. Li Yuan calls this removing the “wall” of perception.
He studied light; the more he studied, the clearer it became. The boundaries of light are much greater than imagined, and there is an increasing number of things he doesn’t understand.
“The process of truly getting to know something makes me very happy.”
This sentence may be a clue to understanding why Li Yuan and Beixing bet on physical AI over the past ten years.
The following is the actual dialogue record. The Leifeng Net·Zuo Lin You Li channel has organized and edited it without changing its original meaning.
01 A class by the bedside
Zuo Lin You Li Channel: Back then, you chose the optoelectronics major because of the statement “the 21st century is an era of optics.” You continued your studies until you got a doctorate. What attracted you to “optics”? Looking back today, has the optical era you imagined actually arrived?
Li Yuan: The term “optical era” was used to refer to a period relative to the electronic era. Light travels faster and has higher frequencies.
Over the years of studying, doing a doctorate, and working as a postdoc, I found that the boundaries of light are much broader than I had thought at that time. There is more and more that I don’t understand, and it becomes increasingly fascinating.
Several areas in the field of technology have been very active over the years: optical communication, optical computing, and even current optical perception—basically everything revolves around light.
Light, as the next medium after electrons, has excellent properties such as high frequency. It will remain an important direction for technology in the present and for a long time to come. It still attracts me.
Zuo Lin You Li Channel: During your postgraduate studies, you studied under Academician Xu Xùróu. What was the most influential thing Academician Xu had on you?
Li Yuan: The inspiration given to me by Academician Xu is not only in my papers, but also in my perspectives. I’ll talk about two things that I find interesting.
The first thing is that on the walls of general laboratories, slogans such as “Diligent and Serious” and “Brave in Innovation” are usually posted. However, our laboratory has eight big characters: “Patriotism and Honor of the School, Never Forget National Humiliation”.
Academician Xu experienced the War of Resistance Against Japanese Aggression. He traveled from Shandong all the way to the Southwest Associated University to pursue his undergraduate studies. He was a young person of that generation who carried strong feelings of loyalty to his country. Later, when we went abroad to earn a doctorate, he was an active supporter. He believed that if he could study in the Soviet Union, we should also explore foreign countries. He truly thought that scientific research and a sense of loyalty to the country were interconnected.
Second thing: In his later years, he often had minor health issues and needed surgical treatment. At that time, I organized my classmates to take care of him. Once it was my turn, he lay on the hospital bed and chatted with me: “Li Yuan, as a researcher, whether you win the Nobel Prize or become an academic or professor, what determines the height of your life is not how much knowledge you possess, but what your worldview is like. The greater your worldview, the more likely you are to achieve great results.”
These are two things that left the deepest impression on me from his guidance. First, it made me realize that scientific research has its own sense of national responsibility. Second, developing an understanding of the physical world and a worldview is more important than being able to derive a formula.
This influenced my future path. I will consider the boundaries of technology, think about what technology can bring about changes for the entire world, and be willing to push new technologies in more directions, rather than just focusing on making high profits and holding things in one’s own hands.
Zuo Lin You Li Channel: After obtaining a doctorate, you already have a clear path in scientific research. Why didn’t you continue to be a scholar, but instead choose to return to China to start a business?
Li Yuan: First of all, I still really want to do academia today, and I still love the path of scientific research.
Secondly, almost all researchers have a desire for their work to be used by more people and to benefit more people. I chose to start a business because I also hope that the things I develop can spread more quickly. This was the path I found at that stage.
Now, I also spend a lot of time each week reading the latest research and learning new things. Although very busy, we still published a paper on next-generation perception in May this year.
02 The lighthouse of the park, and the homework outside the laboratory
Zuo Lin You Li Channel: After returning to China, you started a business in Nanjing with a few friends. Looking back now, what is the most vivid image you have?
Li Yuan: At that time, we claimed to be “the lighthouse of the park”. Because we arrived at work first, everyone in the entire park could see our lights on nightly, just like a lighthouse in a park.
Zuo Lin You Li Channel: Who is the first customer who is willing to pay for you?
Li Yuan: Zero Degree Intelligent Control was once known as “North Zero Degree, South DJI”. The CEO is Yang Jianjun, and I really admire him. His understanding of drone products is, in today’s view, advanced and highly valuable.
Zuo Lin You Li Channel: How many problems did you encounter at first when bringing hardware to the customer’s site from the laboratory?
Li Yuan: Essentially, it’s a lack of capacity to handle a complex world.
I used to think I was very strong academically, good at innovation and problem-solving. However, the manufacturing industry has its own knowledge system, as does business. At that time, I didn’t have it.
From technological innovation, to turning innovation into products, and then into goods for mass distribution, 70% of the understanding is missing—and this is the essence of the problem.
So it’s normal for problems to occur; there is no shortcut. Just like learning physics, you need to read the books you should read, derive the formulas you should do, and memorize what you need to remember.
At the beginning, there were problems with the yield rate, and there were also issues with delivery capacity. Today, the defect rate is basically controlled at a few millionths level, and delivery capacity has improved; nearly a million units are delivered per year.
Zuo Lin You Li Channel: When the organization was small, the founder could focus a lot of energy on research and technology. As it grows larger, there are more management and business-related tasks. When did you realize that you must become a CEO?
Li Yuan: I have always been curious and interested in this world, and I always approach things with the mindset of a researcher. Whether conducting research, managing supply chains, developing products, or working in marketing, I always do it with the idea of “discovering patterns and applying them.”
A technology becomes a product, and the product must be stable and producible at scale; then it becomes a commodity, which needs to be tradable and purchased by everyone. As we move further forward, the methodologies, worldviews, and tools change differently. For us, it’s a process of relearning and rebuilding everything.
I will try to look at the latter part of the matter just like observing pure physical research. It also has its own structure, cause-and-effect relationships, and process. If you respect the laws and do it well, it will work properly.
Over the years, Beixin’s products have all aspired to be leaders in their field. We have carefully considered the needs and challenges of customers in this industry. We won’t see Beixin copying other products; the TF series and Horn series are the same. Essentially, we still use the thinking of a learner during our student days to understand the world.
Zuo Lin You Li Channel: To this day, do you feel that understanding human nature and business is more difficult than understanding light or technology?
Li Yuan: From a mathematical perspective, it’s definitely harder than before.
But in terms of the true complexity of the system, understanding business and sales is not any simpler than learning physics. It has many variables, long processes, and high levels of uncertainty. In business, one is always balancing certainty and uncertainty.
So, back to your question, it’s all quite difficult, but it’s also quite interesting.
03 “Doing well for the robot’s eyes” has not changed
Zuo Lin You Li Channel: The company has been established for 11 years. From the beginning with drones and industrial applications to today, how many strategic turns do you think Bei Xing has gone through? Why did it start with drones and industry instead of vehicle-based applications?
Li Yuan: We have always been obsessed with creating good robot eyes. Whether they are in the sky, on the ground, in water, or on the road, expanding the different dimensions of the physical world is something we are very willing to do and have always persisted in doing.
At the end of 2015, I attended a meeting where someone said: We want to create a product that can float to every corner of the world, from Mars to rural areas in Africa. We hope many people will use it. We are still working on this today.
Around 2021, when starting to develop automotive radar, a question was asked about why we decided to make automotive radar. I said that what we always wanted to do well is to create “eyes that can perceive the world”. It’s not about making automotive radar; it’s about creating excellent eyes. The automotive industry requires these eyes to have reliability and safety, and to truly advance in terms of technology.
Looking back today, the fields of drones, vehicles, and robots together helped us, which led to a significant increase in sales this year compared to last year.
Zuo Lin You Li Channel: Among these tracks, which ones are actively chosen, and which ones are driven by practical constraints?
Li Yuan: The saying “heroes are created by the times”. Keeping our eyes sharp is what we have always adhered to, but we also need to consider the value of the specific scene at a certain stage.
When we first started our business, drones had many challenges. How to land and take off smoothly, fly close to the ground, and avoid obstacles – so we focused a lot of our efforts on creating the “eyes” for the drones.
Later, with the development of China’s railway industry, we have achieved a certain scale in the railway sector. This area continues to develop today, including the science and technology projects we undertake and the upgrades of next-generation products.
Go to the vehicle section, increase the wire count to 512 wires, and then look at the underwater and embodied products released this year. Each product considers the needs and pain points in this field, rather than just being a follower.
TF, Horn, NOVA, UW series, including the new series to be released at the end of the year, we have been answering various types of questions: sometimes railway-related questions, sometimes drone-related questions, and sometimes vehicle-related questions.
On the other hand, solving these problems has also given us capabilities. Drones have enabled us to pass the DO-160G aviation standard, allowing us to work in many demanding environments; vehicle-mounted systems have greatly improved quality assurance; underwater systems have filled the gaps in waterproofing and underwater optics knowledge. Each field has its own system of knowledge, and we need to learn it.
Today, China’s technology is no longer just “doing what the West does” 40 years ago. Many areas have already started to lead the way. Therefore, one must think about the challenges in the industry on their own, be brave enough to lead, and dare to define things.
Zuo Lin You Li Channel: Could you take a specific product and explain it in simpler terms? How did you solve these problems?
Li Yuan: Take the newly released UW500 this year. It is used under water. The speed of light under water is different, there are different nonlinear optics, and the refractive index also changes. The entire optical system is completely different.
To enable underwater robots to perceive their environment, we have conducted many new underwater optical studies. At the same time, we need to make power consumption, size, and cost levels that are compatible with consumer-grade robots. Finally, the radar must be able to clearly see objects and understand the environment underwater, so that the robot can carry out its tasks.In terms of underwater radar, previous giants were overseas companies; I believe Chinese companies have gained an advantage here.
The second is rail transportation. Our Horn radar is still being upgraded this year, and a new railway radar will be released in the future.
The third is the NOVA20 released this year, designed for embodied intelligence products. Its core concept is “slow thinking in the brain, quick perception in the cerebellum.” In the face of collisions and bumps, robots don’t have time for full seconds-long thinking; they need to avoid obstacles and maintain safe operation at the millisecond level.
These products are developed by first considering what specific problems exist in this field, and then working on solutions based on those issues.
Zuo Lin You Li Channel: For laser radar companies, is the robot market a real new growth area, or the second battlefield after intensified competition in the automotive sector?
Li Yuan:Bei Xing is a Robot Native, a native robot lidar company. From the first day, what we wanted was to create the eyes for robots.
Whether it's drones flying in the sky, robots in the water, or robots on roads and in courtyards at home, there are differences. But the design of the eyes has never changed.
For Bei Xing, vehicle-based orientation is a transfer of capabilities. Today, high-level robots have presented us with a new challenge, and we are willing to answer it well. Currently, robots are the main battlefield for Bei Xing.
Zuo Lin You Li Channel: Ten years in entrepreneurship, many new opportunities and trends. How do you decide whether a direction is worth continuing with or if it’s time to cut losses?
Li Yuan: The things we do can be divided into two categories: one is exploring the boundaries of capabilities, and the other is creating economic value.
If something at this stage can push forward the boundaries of human capabilities and technological capabilities, we consider it valuable. If it can bring value to customers and lead to commercialization, we also consider it valuable.
When does it end? When both values are gone.
Zuo Lin You Li Channel: Is there ever a time when giving up was more difficult than persisting?
Li Yuan: Fortunately, the fields we are in are sunrise industries. We have not yet gone through a complete cycle of technological decline, so there has been no moment when “we must give up this field.”
We have been working on drones for 10 years, and today, they remain a hot topic in the low-altitude economy. Last year, we also released a drone product with a frequency of 10,000 hertz. Two years ago, we launched a drone weighing only 10 grams that meets DO-160G standards. There is still much technical advantage left to be explored, and it cannot be said that we have stopped.
The railway sector is the same. China’s railway system is already very strong, but in terms of perception, we are still developing new products.
04 On-vehicle is a necessary hurdle, and robots are the main battlefield
Zuo Lin You Li Channel: When deciding to enter the passenger vehicle market in 2021, why did you insist on entering?
Li Yuan: At that time, our understanding was that if the hardware and the design are good, and the vehicle is essential for success, the quality control, processes, and environmental adaptability of the vehicle system greatly improve the product. Therefore, this field must be developed.
Today, it seems that many of our security measures were actually implemented at that time. Now, when performing remote upgrades for robot customers, we still use the capabilities that were developed during that period.
Zuo Lin You Li Channel: In 2023, there were no actual shipments for the in-car installation business, but a lot of research and development efforts were put in. Was that the stage when you started your business where you doubted yourself the most?
Li Yuan: In terms of our expectations, to do well in radar, we need to learn from this field and improve our skills. Therefore, we won’t be swayed by short-term achievements and affect our judgment on the direction.
Later, the vehicle business also brought us mass production and orders. Objectively speaking, the vehicle industry is helpful for our radar.
Zuo Lin You Li Channel: Today, Bei Xing placed the robot in a higher position. Is this a return to the starting point of entrepreneurship, or have we entered a new stage?
Li Yuan: What we always want to be is “the eyes of understanding the world”.
Today, robots have become the interface for interaction between the physical world and the digital world. The new question is how to connect AI with the physical world, allow AI to move beyond software and interact with the real world, and even perform certain actions.
The robot carries us forward in this step from lidar to physical intelligence.
Zuo Lin You Li Channel: Do you think the robot market has reached a point where actual volume starts to increase?
Li Yuan: It can be divided into two parts. One is the construction of AI infrastructure. Although it has been built quickly, it is not yet fully completed. The other part is the commercialization and productization based on this infrastructure. Now, the application side has already started.
The future of robots has a vast scope. Looking back at today from the perspective of the next 10 years, robots are just at the beginning.
Today, we might start with consumer robots. Next, embodied intelligence will begin to be widely used in practice. When people around you or in the office start having embodied robots or robot dogs, I think the penetration rate will really increase. It’s still just the beginning.
Zuo Lin You Li Channel: Where are the new products and technological upgrades in the railway sector mainly reflected?
Li Yuan: The specific details will be revealed later. A clear change is that customers have moved from the early phase where “just having a feature was enough” to now demanding a more humanized design with a sense of comfort.
When installed on the train, it must not only function properly, but also be small enough and have low power consumption, and be quiet so that the driver feels comfortable.
China’s railways are beginning to lead the world. As a leader, they can’t just focus on functional implementation like in the past. Issues such as temperature control and noise—which were once considered tolerable—can no longer be tolerated now.
Zuo Lin You Li Channel: How much of the business world can be quantified? How much depends solely on understanding people?
Li Yuan: Objectively speaking, I am not a particularly outstanding salesperson; my background is still that of a researcher.
Over the years, our market share in overseas markets has been increasing, and our products are becoming more accepted by European and American customers. This is due to our growing understanding of the customers, the local geographical environment, and how to provide excellent service.
The so-called understanding of geography means knowing how a product should function in extremely cold or extremely hot regions. You can’t just say the product is good at home; when placed in that area, it must really work well.
The so-called good service doesn’t just mean building good relationships. For example, setting up agents in the local area; even if I can’t go there today, the agent can come to fix things this afternoon.
Our team, from quality and procurement to senior management, works in one-on-one relationships with customers. When issues arise, they can find us immediately. We also respect the customer’s processes and carry out testing, negotiations, EVT, DVT, and PVT according to the product procedures of their country, until mass production begins.
This is actually a quite engineering-like approach. Over the years, what I really improved is my understanding of the country’s technology, processes, and organization. This is the process we technical people go through when learning about business and expanding overseas.
Zuo Lin You Li Channel: What is the approximate proportion now both overseas and domestically? Is there a difference in the markets for shipments overseas and domestically?
Li Yuan: The overseas proportion currently exceeds 50%, and this has been the case for many years. In terms of industry, there are no boundaries in technology; both domestic and foreign markets rely on robot-based customer bases.
05 From “what can be done” to “what can be brought to the customer”
Zuo Lin You Li Channel: Compared to three or five years ago, what is the real change that has occurred in the robot market? What has changed in the logic of customers purchasing lidar?
Li Yuan: In two aspects: the boundaries of capabilities, and the value it brings to the world.
The capacity boundaries are what you can actually do. How many lines of radar, how high the resolution, and what quality requirements are met. The value lies in how much the customer’s accident rate can be reduced and how much the cleaning efficiency can be improved.
The earlier a field is, the more emphasis is placed on capacity boundaries; the more mature it is, the more value is discussed.
For example, in embodied intelligence, we are still working hard to define clear boundaries for its capabilities. NOVA20 can complete the entire detection process within dozens to hundreds of milliseconds, ensuring that sunlight does not interfere and signals do not cause interference. This is performing capabilities that were previously impossible.
With technological advancements, the discussion has shifted to what value robots bring to customers in specific scenarios.
The same is true for drones and the low-altitude economy. In the early days, radar was provided to logistics drone companies, and more focus was placed on exploring the boundaries of capabilities: passing the DO-160G, reducing the radar weight from over 100 grams to 10 grams.
Currently, we are working on this year’s modification for the customer. The focus is on improving supply chain capacity and reducing power consumption, so that the flight range can be further enhanced.
This is a transformation over the years. In some fields, the logic of capability boundaries has been largely completed, and there is a shift towards value logic.
Zuo Lin You Li Channel: Which changes are substantial? Which remain at the capital and dissemination level?
Li Yuan: Let’s be straightforward and simple:
Any internal competition centered around performance-driven activities is, in essence, still exploring the boundaries of capabilities. When it starts to actually deliver results, discussions begin about the value for the client base.
From the demo performances, to actual work, and then to discussing whether the cost is low enough and if the customer can afford it, one can gradually feel the evolution stage of an industry.
Zuo Lin You Li Channel: The robots start working and engage in actual production activities. Does this mean that the industry has reached a critical point?
Li Yuan: Yes. Because once it creates value, that value will roll back to the initial state, becoming a flywheel.
The previous money was only used for research and development investment. The more money made, the more can be invested in research and development; the more research and development is invested, the more money is made. When this cycle starts spinning, it enters the second golden period.
Zuo Lin You Li Channel: In fields such as humanoids, courtyards, pools, drones, warehousing, and industrial robots, which one is likely to form a large-scale, sustainable market first?
Li Yuan: Look at it from two dimensions.
On the intelligent boundary, our humanoid robot customers and some high-intelligence robot customers are working together to explore the upper limits of perception.
When it comes to actual value, it depends on the quantity.
At different stages, the features that a product exhibits are different. Some show increased capacity, while others show “things I could not do before,” such as running and jumping.
With implementation as the core, focus on scalability; with exploring the boundaries of intelligence as the core, look at the upper limit of perception.
Zuo Lin You Li Channel: In today’s robotics industry, which scenarios are overrated, and which are underrated?
Li Yuan: I can’t say whether it’s over- or under-estimated, but I can share some numbers for reference.
According to Zuo Shi Consulting, the global pool cleaning robot market size will grow from $1.394 billion in 2020 to $2.8 billion in 2025, with a compound annual growth rate of 15%. It is expected to increase further to $4.568 billion by 2030; the market penetration rate will rise from 27.8% in 2025 to 34.7% in 2030.
Looking at IDC’s “Global Home Smart Cleaning Robot Market Tracking Report,” the total global shipments of home cleaning robots reached 32.72 million units in 2025, a 20.1% increase year-on-year.
Among them, in 2025, the global shipment of lawnmowing robots reached 1.992 million units, a 63.8% increase year-on-year, making it the fastest-growing segment.
Zuo Lin You Li Channel: The vehicle industry later moved towards price wars and consolidation. Will the robot laser radar follow this path?
Li Yuan: My basic view is still value logic.
A low price is not a bad thing. Only when the price is low can more people benefit from technological advantages. If a bicycle still costs a month's salary, I simply can't afford to ride it.
But zero or negative gross margins require reflection. When you create something that doesn’t bring value to the customer, pricing becomes a losing proposition. What really needs to be discussed is whether we bring value to robot users.
I have been working with drones for 10 years, and I can still make money from them. I can also upgrade them every year. The same is true for robots—only when you add value to them will they continue to upgrade and grow.
The price of a single product needs to be reduced. The gross margin cannot be as high as before, but it is still possible to make a profit. As the total gross margin increases, research and development can continue.
We have always emphasized value logic because we have actually “eaten meat” from it. In the past 10 years, fields such as drones and railways have allowed us to make money. Then we invest that money in research and development to create new products, and once again, we can “eat meat”.
We are a company that has emerged from the intense competition in vehicle industry, so we cherish value even more. Don’t blame the era or the industry; if you have value, customers will be willing to give you higher margins.
Zuo Lin You Li Channel: Companies like HRS and Sutek also are expanding their robot laser radar business. What is the difference between Beixing?
Li Yuan: The robot market has many specialized categories. You can see some of the products we have developed.
For drones, TF products still hold a leading position today, and their adoption rate is continuing to increase; for rail transit, the Horn radar is still being upgraded; and for underwater robots, UW and VLS have an increasing market share.
For embodied intelligence, although NOVA was just launched, I believe that when we chat again next year, you will see that many users have already started using it to reduce collision rates and accident rates.
Still back to a very simple question: what value do you bring to each specific industry?
BeiXing is a company of Robot Native, which is accustomed to considering issues from the needs of the robots themselves. This is what has been accumulated over the past 10 years.
Zuo Lin You Li Channel: Is getting in touch with these robot customers earlier a very important advantage for Bei Xing?
Li Yuan: Besides morning and evening, there are also more intimate and deeper interactions.
Following is one way of thinking, while leading is another way of thinking. The essence of leading is to select the right niche market, get close to it, and integrate with the customer value of that niche market.
There will be a long climb during this process, but once you get past it, you can gain technical feedback and grow from this field.
Our principle is to select the right niche market, provide excellent robot services there, and then define what kind of eyes it needs.
06 Twenty points, is that enough?
Zuo Lin You Li Channel: Use three numbers to describe the progress of Beixin in the robot market. Which three would you choose?
Li Yuan: In the first half of this year, we delivered 250,000 radar units to robots; robot revenue accounted for over 50% of Beixing’s total revenue; in 2025, robot revenue will increase by 100% compared to 2024.
These numbers are still increasing.
Zuo Lin You Li Channel: NOVA20 chose a 20-point line array, without pursuing a surface array solution with a higher beam width. How do you decide when to aim for high parameters and when to make cuts for the scene?
Li Yuan: Essentially, it is our understanding of embodied intelligence: "The brain thinks slowly, while the cerebellum senses quickly."
First, it depends on whether perception serves the brain or the cerebellum. NOVA20 serves the cerebellum, focusing on security and rapid perception, and must be highly integrated with the motion system.
These 20 points are activated one by one. To prevent them from interfering with each other, we divided the time and conducted point-by-point testing. This technique is quite effective; it was previously used in vehicles, and we have brought it over to apply it to such a small product.
Our other product philosophy is called “Simplicity Over Complexity”.
Select 20 points because near-field requirements require faster scanning and faster results. Meeting the customer’s needs truly is more important than stacking parameters.
Even though there are only 20 points, its blind area is very small, just a few centimeters. It can still be used under strong outdoor light. The size is only as big as a beer cap, and it can fit in many parts of a robot body—the elbow, hand, leg, head, and chest.
The cost is also consumer-grade, and it is designed from the start for large-scale deployment.
Zuo Lin You Li Channel: These 20 points are sent out sequentially with a time difference. What is the difference between them in a specific scenario?
Li Yuan: This is time-division multiplexing. Each point is transmitted and received independently, with a very short delay in time—perhaps just a few ten-thousandths of a second—but it is sent in a time-division manner.
Previously, lidar had a major problem: if there was something extremely bright outside, it would "light up" the other points around the excitation point and contaminate them. This is called crosstalk, which has troubled the industry in vehicle radar for many years.
Later, everyone used time-division multiplexing to solve it. This technology was previously used only by expensive radars that cost thousands or tens of thousands of yuan.
We have passed the vehicle-mounted stage and produced a large amount of output. Now we apply it to the embodiment section to solve the crosstalk problem between perception points.
Zuo Lin You Li Channel: The technical approach of Beiqing is mainly solid-state and non-mechanical. Why do they choose mechanical rotation in underwater scenarios?
Li Yuan: It is determined by the customer’s requirements, the underwater environment, and costs.
During the project initiation, we evaluated many options, such as using multiple solid-state components to form a circle. Finally, after discussing with the client, we considered both the ease of installation and how well the client’s algorithm worked. Their previous algorithms were based on mechanical rotation-based SLAM.
Either have the customer rework a lot of things, or in this form, enable the customer to use it faster and better. We prioritize the latter, and finally we adopted a rotating chassis to achieve 360-degree scanning.
I don’t think there is any conflict. Being loyal to a technical approach is a obsession; being loyal to physical laws and customer value is our philosophy in developing products.
Zuo Lin You Li Channel: You previously said that lidar will eventually move towards semiconductorization. When was this judgment formed? Which discrete capabilities have been integrated into chips so far?
Li Yuan: As for the chip part, I will provide a detailed explanation to everyone next time.
One thing to share is that the essence of hardware development is to gradually chip it up, because chipding is most suitable for mass production on a super-large scale.
Zuo Lin You Li Channel: In the past, Bei Xing provided distance information. Now, people are starting to discuss perception models and environmental understanding. What has changed in your understanding of “machine eyes”? As the brain becomes smarter, can inexpensive cameras also solve the problems that required lidar in the past?
Li Yuan: Eyes are not just tools for the brain, but also a source of knowledge.
Cameras and lidar are not opposing technologies. Cameras are passive sensors that capture RGB color information, providing rich data with large volumes of information; lidar is an active sensor that directly measures physical quantities such as distance, position, and reflectivity.
In the past, discussions focused more on the capabilities of individual sensors. In the era of large models, cameras provide rich semantic information, while lidar offers more precise measurements of the physical world. The combination of both is an inevitable trend. We will also share good news in this area soon.
Zuo Lin You Li Channel: Looking further ahead, in the era of robots and physical AI, is there any judgment in an industry where a consensus has not yet been reached, but you are quite confident about?
Li Yuan: In the end, we need to use lidar to enhance the model’s capabilities.
Lidar is a excellent physical truth expressor. With the basic model combined with lidar, through multimodal fusion, results can be far superior to models trained solely with cameras. The paper we published in May this year also discusses this direction.
The biggest benefit of this era is the benefit of large models. As a physical truth device, lidar radar has great value to be realized in the era of physical intelligence.
Zuo Lin You Li Channel: “Physical AI” has been frequently discussed recently, but Bei Xing’s “eye creation” has been going on for 10 years. Is this a real paradigm shift, or are we just giving a new name to an old problem that has been studied for a long time?
Li Yuan: I can only share my feelings.
AI is like a silly, confused child with endless desires for the world. After several years of exploring and expanding in the software world, it starts to wonder if it can step out of software and explore the hardware and the real physical world?
The current essence of AI is that it is not willing to exist only in the virtual world; it begins to create itself in the physical world.
So there will be AI for Science and Physical AI. Next, there will be AI for Engineering, including the construction of AI infrastructure. The core idea is to emerge from the virtual world. We also want to make some innovations in this trend.
Zuo Lin You Li Channel: Imagining the future 10 years, what problem do you most want to solve? What kind of company does Hope Beixin hope to become in the end?
Li Yuan: If I can play a core role in the process of “AI creating itself”, I would be quite happy. Right now, we mainly act as eyes and sensors.
Making AI truly exist in the physical world, knowing how to create itself and how to interact in that world, is something that excites me greatly.
07 The boundaries of the company, the background of researchers
Zuo Lin You Li Channel: We talked a lot about Bei Xing’s strengths earlier. Are there any obvious weaknesses?
Li Yuan: Yes, there is a lack of investment in markets and promotion.
We should actively communicate with the media earlier, instead of starting conversations today. Learning how to promote, understanding promotion, and using promotion tools effectively is a lesson that needs to be learned next.
Zuo Lin You Li Channel: Bei Xing covers four directions—"air, land, water, and ground." Only one direction can be deeply involved. Which one will you choose?
Li Yuan: Let me first explain why we have so many “Airland Waterland” products today.
If the company has only one day, we won’t be working in four directions simultaneously. But over the past 11 years, we have gone through many economic cycles and ups and downs. We have gotten back on our feet, improved customer service in these fields, developed products to this point, and managed to make a profit.
I have brought value to the customer, and the customer has given me enough gross profit. The product is still being upgraded, and I will not give up these areas.
When will I give up? When I feel that I can't add value to a field, I will stop and hand it over to companies that are better at operating during the saturated and convergent periods of the industry.
It can be broken down into parts. Some are existing businesses that have reached a mature business stage. For me, it means doing business well, selling products well, and providing good service.
In terms of technological innovation, we focus on robots. How to make eyes more intelligent, and how to improve AI’s perception of the physical world, are the areas we spend the most effort on research in now.
Zuo Lin You Li Channel: Currently, chip companies produce radars, radar companies produce chips, and robot brands are also developing their own sensors and algorithms. Is this a temporary phenomenon in the early stages of the industry, or will vertical integration persist in the long term? Where should the boundaries between these entities be?
Li Yuan: The fact that everyone is still competing for space indicates that this field is still developing. If there is no more competition, it means it has little value.
I believe vertical integration in the field of robotics will persist for a long time. Each stage has value that can be contested, and there is no pre-established reasonable division of labor. Whoever can create value for customers will expand their boundaries, and eventually a balance will be reached.
Take the mobile phone camera as an example. In the early days, everyone tried to gain an advantage in the user experience. Later, Sony and Samsung focused on sensors, while Apple and Huawei took over the ISP and imaging algorithms. The division of labor became clearer: specialized manufacturers handled components, and OEMs were responsible for the user experience.
Devices are a large-scale business; they need to serve the entire industry by spreading the manufacturing costs over billions of units. Experience, on the other hand, is a business related to defining rights, and the complete device manufacturers must hold those rights in their own hands.
Robots may move towards a similar division of labor: chip manufacturers produce general-purpose platform components, sensor manufacturers produce physical components, and they also extend slightly to perception algorithms; robot manufacturers focus on overall system definition, task layers, and intelligence.
But it’s still early; vertical integration and specialized division of labor are still operating simultaneously, and the final product may not necessarily be a complete copy of the phone.
Zuo Lin You Li Channel: From a economic perspective, how much production volume can a pool robot company achieve with in-house laser radar?
Li Yuan: Unless a robot company occupies 80% or 90% of the market, there is actually no need for in-house development.
Even if a large company holds 20%, the self-developed products can only be sold in that 20% quantity. A component factory can serve 60%–80% of customers. As long as the quality and cost are good, it’s sufficient.
The essence of division of labor is still value. Who occupies a higher position in the value part, that person gets the money. The pool pool plant makes its own components, and the value logic doesn’t hold easily.
Just like many car manufacturers said they would develop their own lidar back then, is it really worth it in the end?
Radar is essentially a robot’s eyes. It is still in development, constantly adapting to customer needs, continuously upgrading, and creating next-generation products.
The car factory later gave up, one reason being that the iteration speed was not fast enough. The best talent was responsible for defining the complete vehicle and component specifications, but their speed wasn’t that fast. To maintain advancement, it was necessary to buy products from independent manufacturers.
Unless this thing is no longer being upgraded, then developing it in-house isn’t much of a solution. But such fields aren’t suitable for an innovative company like ours.
Zuo Lin You Li Channel: Among your entrepreneurial experiences over the years, what was the most correct decision you made? What was the highest cost you incurred?
Li Yuan: The most correct judgment is the commitment we have had over the past ten years to robots. By persistently developing products across the entire spectrum, and laying a solid foundation in perception step by step, we can reap the technical benefits today.
If there is a cost, it is youth.
That day when I looked in the mirror, I suddenly realized that I had a lot of white hair. A few days ago, when I returned to my alma mater for the 130th anniversary event, I found that many teachers and classmates also had white hair.
We researchers really enjoy what we do, but time passes too quickly.
In the next 10 years, I hope to value my time more and use it in technological innovation and business implementation, bringing some color to this era.
Zuo Lin You Li Channel: Some people say you are rational and calm, and good at analysis. Do you agree? Is this the important reason that supported your business venture over the years?
Li Yuan: This evaluation is too high; I’m a little embarrassed.
When it comes to underlying tags, first of all, I am a Beijing person with a strong curiosity. A major characteristic of Beijing people is their love for “birdcages and perches”, but I simply don’t have the time for that. On the other hand, by working on every product and spending so much time in the laboratory every day, isn’t that also satisfying my own curiosity?
Second, I am a “good student”. I prepare my papers carefully, do my homework diligently, and defend my projects seriously. For my undergraduate, master’s, and doctoral theses, every word was written by myself.
Today is the same. I have a strong curiosity and a deep obsession with technology. It’s like when I was studying for a doctorate—sitting in a chair, not getting up for a whole month. I think that might be my fundamental nature.
Zuo Lin You Li Channel: As mentioned in previous reports, you deliberately challenge something you dislike every year. Why train yourself like this?
Li Yuan: I always feel that whether one likes something or not is often related to some wrong perceptions in the mind.
For example, if you don’t like a certain subject, it might be because you are afraid of it, or perhaps you don’t know enough about it. Over the years, my biggest realization is that any subject, anything—no matter whether you liked it at first or not—as long as you start learning and immerse yourself in it, you will eventually find it quite enjoyable.
So, I go to challenge some fields that I used to be reluctant to do or like before every year. Essentially, it’s about breaking my own cognitive traps, or cognitive barriers.
I am very happy that I can truly get to know something.
Written at the end
In the original interview stenotype, Li Yuan mentioned “value” a total of 55 times.
This word is very common in business conversations. The phrase “create value” placed at the end of a sentence sounds like the reason for doing something has already been clearly stated.
However, when “value” is followed by specific things, its meaning returns.
When talking about the next generation of railway radar, Li Yuan mentioned a requirement: quietness.
In the past, when radars were installed on trains, customers might accept them as long as they could perform their functions. Now, they also have to be small enough and energy-efficient, with proper temperature control and noise reduction. To make the driver comfortable, those issues that could be "tolerated" in the past can no longer be tolerated today.
Create eyes for machines, and eventually, even human ears must be taken into account.
Many years ago, Academician Xu Xùn said to Li Yuan from his hospital bed that what determines the height of a scientific researcher’s life is their worldview. After reading this passage again, I pay even more attention to that “world” — it includes physical laws yet to be explored, the yield rate in factories, the bills in customers' hands, and a driver’s perception of noise.
A person’s time is limited, but the world they can know is still expanding. Speaking of this process, Li Yuan said, “I’m very happy.”
We welcome all group members who can provide new perspectives, new angles, and new information for the Leifeng Net (Public Account: Leifeng Net) · Zuo Lin You Li Channel. Join our real-name community and walk together with Zuo Lin You Li. Administrator WeChat ID: leiphonelinli, note: company - position - name via faster communication.
Leifeng Net original article, unauthorized reproduction is prohibited. For details, see Reproduction Guidelines.