Interview with Li Tong, CEO of Keenon Robotics: Don't sit at home and imagine when it comes to robot commercialization.
Text by | Qiu Xiaofen
Edited by | Su Jianxun
During the exhibition season in July and August, if you visited Keenon Robotics' booth, you would have seen simulated scenarios such as bars, restaurants, and theaters. Their bipedal service robot, XMAN-F1, transformed into a waiter, helping visitors make popcorn, prepare iced drinks, mix cocktails, and so on.
The idea of having robots do real work - this setup concept stems from the business philosophy of Li Tong, the CEO of Keenon Robotics. When communicating with Intelligent Emergence, Li Tong, the CEO of Keenon Robotics, repeatedly mentioned the word "commercialization".
As a commercial robot company that has weathered the ups and downs for 15 years, Li Tong believes that this is precisely the key word for their survival and "crossing the cycle" in the previous round of the commercial robot "escape".
△ Keenon Robotics' robot shoveling popcorn at the WRC site. Photo source: Shot by Intelligent Emergence
However, it is easier said than done to commercialize robots. Li Tong said that for robot implementation, one cannot just "imagine" what customers need while staying at home. Instead, one should bravely go to the site, observe the real pain points of customers, and then find the intersection between the robot's capabilities and customer value.
"There is a gap between what the client expects and what you imagine, and this gap needs to be bridged."
Li Tong gave an example to Intelligent Emergence. When their robots were deployed in a certain chain hot pot restaurant, their R & D personnel, product managers, and project managers all went to the site. Even when the robots had problems, they even helped deliver dishes.
In order to improve the granularity of robot services, they need to understand every small detail of the customers, down to the customer's business processes, the responsibilities of each job position, how a new employee gets familiar with the work from scratch, the customer's training content, standards, and so on.
Li Tong said that taking this hot pot restaurant as an example, there are more than 20 types of jobs in a store. Among them, there are part - time and full - time dish delivery staff. Each person needs to use a bracelet to swipe a card for piece - rate work when delivering a tray, and the payment for delivering one tray is 50 cents.
However, how to judge whether the commercialization of robots is successful? Li Tong put forward the concept of "job - based" robots. In his opinion, when a robot can truly replace the work of a job position and become a more efficient labor force than humans, there is a possibility of large - scale implementation. Otherwise, the so - called robot work is just a display of skills and a gimmick.
He said that after calculation, Keenon's robots can basically replace the position of a human employee, and the cost is 1/2 - 1/3 of that of a human. "A robot company is essentially a labor force company. The key to commercialization lies in whether the robot can become a labor force."
Currently, Keenon has sold more than 100,000 commercial robots. A 2024 IDC report shows that they are the world's number one in the shipment volume of commercial service robots, with a market share of 23%.
The robots you see delivering goods back and forth in Burger King and Hilton Hotels are all from Keenon. In addition, their product portfolio also includes cleaning robots, which have been exported to more than 60 countries and regions around the world, including Europe, America, Japan, and South Korea.
△ Keenon's luggage robot. Photo source: Official
When the shipment of their specialized robots became gradually stable, starting from 2023, Keenon also began to develop the "brain" of robots, and the product concept shifted from specialized robots to general - purpose humanoid robots.
Li Tong described such a development path to us: Only when robots have large - scale commercial implementation in various "fragmented markets" and accumulate enough data, and as the data and models continue to iterate, is there a possibility of intelligent emergence, and a smarter and more generalized robot brain can evolve.
During the WRC, we communicated with Li Tong about the secrets of their commercialization, how to transform robots from specialized to general - purpose, and their experiences in exporting robots overseas. Perhaps it can provide some reference value for the new wave of robot startups.
The following is the transcript of the communication between Intelligent Emergence and Li Tong, published after editing:
A robot company is essentially a "labor force" company
Intelligent Emergence: After the wave of embodied intelligent robots became popular, what's your mindset? Compared with many newly emerged companies, what's your competitiveness?
Li Tong: Now that embodied intelligence has emerged, we need to openly embrace this wave. In fact, this wave has only emerged in the past two or three years. Everyone is on an equal starting line. On the contrary, our in - depth understanding of the industry and our dedication to products give us an advantage, enabling new products to be quickly implemented, while some companies are still in the process of showing off their skills. On the other hand, we also know how to go global, understand the cultures of different places, and then design different product forms and products.
Intelligent Emergence: After being in the robot industry for 10 years, how do you feel the difference between the robot industry in the past two years and that in the earlier days?
Li Tong: Now it feels like the era of mass entrepreneurship and innovation. We've been in this industry for 10 years. Basically, we experience a cycle every two or three years, with periods of attention followed by periods of decline, and we're used to it.
We're a company that has survived the cycles. In fact, four or five years ago, there were hardly any service robots in our lives. At that time, there were many people trying to make (service robots), no less than 80 or 100 companies.
We see a similarity in history. The situation then was the same as today. How can so many manufacturers make it to the end? I think from historical experience, only enterprises that truly aim for final commercialization can really survive. In the previous wave, those without commercialization all failed in the end. Currently, the number of large - scale companies can be counted on one hand.
Intelligent Emergence: Everyone knows that commercialization is important. Why do so many companies still fail in the end?
Li Tong: There is a gap between what robot manufacturers imagine as "implementation" and the actual implementation. There is a gap between what the client expects and what you imagine, and this gap needs to be bridged. Do you really understand your customers, or do you just think you do?
For a robot company to truly achieve commercialization, it should be with its customers every day, understand the customer's business processes, know what the robot product can do, understand how much the customer can pay, and finally find the intersection between customer value and product value.
For example, our robots can replace one person, or even more. The price of our robots is 1/2 - 1/3 of the local labor cost. Only in this way is there a possibility of large - scale commercialization.
We put forward the concept of "job - based" robots - when a robot can truly do the work of a job position and is more efficient than humans, there is a possibility of implementation. Otherwise, it can only be a demo and a gimmick.
A robot is essentially a labor force. We joked that a robot company is essentially a labor force company. The key to commercialization is whether the robot can become a labor force.
Intelligent Emergence: Your commercialization is deeply integrated into the customer's scenarios. Can you give an example?
Li Tong: Take a certain chain hot pot restaurant as an example. There are 20 types of jobs in the restaurant. There are two types of delivery staff, part - time and full - time. They are paid 50 cents for each tray they deliver. The delivery staff has a bracelet, and there is a card reader behind it. They swipe the card when they pick up a tray. It's a piece - rate system.
So, the labor cost calculation in this hot pot restaurant is very clear. There is a dish - preparing staff in front of the dish - delivery staff, and a waiter at the end. Just this one process involves three types of jobs.
△ Keenon Robotics' robot mixing cocktails. Photo source: Official
Intelligent Emergence: In the previous wave of the service robot boom, your company stood out in the end. Do you think your competitiveness lies in the most detailed control of commercialization?
Li Tong: We don't dare to say that. But I think for commercialization, one cannot just stay at home and imagine what customers need. Instead, one should bravely go to the site and observe the real pain points of customers.
In the past, we were always in the customer's scenarios, understanding every detail of the customers. This includes the customer's business processes, the responsibilities of each job position, how a new employee gets familiar with the work from scratch, the customer's training content, standards, and so on.
Intelligent Emergence: How is your detailed control of commercialization reflected in organization and management?
Li Tong: Our R & D product managers and project managers all go to the site. If the robot has a problem, you have to go there to help with the delivery and do the work yourself.
After selling 100,000 robots, decide to move from specialized to general - purpose
Intelligent Emergence: Before last year, most of your displays were in the form of hotel service robots. Why are most of them humanoid robots this year? How did this transformation happen within the company?
Li Tong: We've always been thinking about whether to directly develop a general - purpose robot as a big move or to take a step - by - step approach and implement in stages. Finally, we decided that we should first implement in simple vertical markets, which we call segmented markets. Then, while implementing, we generate data.
We believe that the biggest bottleneck for robots now is data. The data in vertical scenarios is vertical and shared. Only through continuous data iteration can we finally achieve a relatively all - around process. This is the commercialization logic we've thought about.
Intelligent Emergence: What are the characteristics of specialized robots and general - purpose robots in different scenarios?
Li Tong: Most of the robots we made before were specialized robots, but this also required us to constantly develop new robots according to different needs.
Actually, both specialized robots and general - purpose robots have their own advantages and disadvantages -
The advantage of specialized robots is that the product is designed specifically for a job position, so it is undoubtedly the most efficient. However, the disadvantage is that it cannot be transferred. If you change the job position, you have to redesign it;
The general - purpose robot is the opposite. Based on the iteration of software and data, it can change to different job positions, but the disadvantage is that the ROI is also low. For example, when delivering trays, a specialized robot can carry four, while a general - purpose humanoid robot with only two hands can only carry two.
In service scenarios, some jobs involve various tasks and are changeable, so general - purpose robots are more suitable, such as waiters. Some jobs only involve one task, such as delivery, so specialized robots are better. For some tasks that can be completed by a robotic arm, I don't think a humanoid robot is necessary.
Only by combining specialized and general - purpose robots can we more efficiently solve the problem of labor shortage in the service industry.
△ Keenon's bipedal embodied service robot XMAN - F1. Photo source: Official
Intelligent Emergence: From specialized robots to general - purpose robots, there are significant changes in the underlying technology and form of the products. How can the experience you've accumulated over the past ten years be applied to the next transformation?
Li Tong: Actually, the underlying technology of specialized robots and general - purpose robots is exactly the same. The electromechanical system is the basic skill of robots. In addition, we do all the motor, drive control, and algorithms ourselves. The only difference between the two types of robots is the embodied intelligent large - scale model, but in this regard, everyone is on the same starting line.
There are various technical solutions for the embodied large - scale model now, and there is no recognized path yet. Wang Xingxing said he is skeptical about VLA, but we still believe in VLA. Maybe he's right, or maybe we're right. But before the large - scale implementation of robots, it's impossible to judge who is right.
Robot startups cannot stay in a vacuum
Intelligent Emergence: Compared with many current manufacturers, one of your advantages is that you've shipped 100,000 robots, have scenarios and data, and don't need to collect data through remote control like many manufacturers?
Li Tong: Yes, but operational data is still needed. The perception of space is part of VLA. If there are 100 robots in a collection and training field collecting data for 8 hours a day, this data is still not enough. So, it still has to be obtained through the large - scale implementation of robots. Without large - scale implementation, where can we get large - scale data? Data and commercialization are a "chicken - and - egg" problem.
Intelligent Emergence: We feel that there is a big gap between the actual performance of robots and users' expectations. What's your feeling about this?
Li Tong: Our hotel delivery robots are standard in China, but they are rarely seen overseas, mainly because of the elevators. In China, it's possible to modify elevators, but in foreign countries, elevators are special equipment, and the cost of modification is very high. So, we thought about adding an arm to the robot to press the elevator buttons, but it's not that simple.
For example, in many hotels, there are 6 elevators. After the robot presses the button, it doesn't know which door will open; the