UBTECH Jiao Jichao: In the next three years, the cost difference between bipedal and wheeled robots will be at most 20%. | 36Kr Exclusive Interview
Author｜Song Wanxin
Editor｜Zheng Huaizhou
In the past two years, when the concept of robots has become a popular topic in the secondary market and the industry has been reporting frequent successes, Youibot, the company that was the earliest to get involved in humanoid robots in China, has become the most low-key one.
At the just-ended 2025 CES, various humanoid robots emerged in large numbers, but Youibot only exhibited an intelligent lawn mower.
Behind this is the rapid strategic shift of Youibot in the past two years. Since 2023, Youibot has identified the B-end intelligent manufacturing scenario as a new strategic focus.
According to Tan Min, Youibot's CBO, the historical mission of Youibot's humanoid robots at CES has been completed, and now the functions displayed by any humanoid robot company at CES have not surpassed what Youibot has previously shown.
Since going public, there have been continuous disputes about Youibot from the outside world. In addition to the expected loss problem as a high-tech enterprise, CEO Zhou Jian's termination of the concerted action with several shareholders also caused the market to panic for a time, leading to a significant drop in the stock price to HK$40.
Recently, after several senior executives of the company disclosed the long-term lock-up of shares announcement, Youibot's stock price gradually recovered to around HK$80.
In the early days, Youibot once targeted bipedal robots for the home service scenario, but Jiao Jichao, Vice President and Dean of the Research Institute of Youibot, said that at the commercialization stage, Youibot found that there are still many problems that cannot be solved.
After that, Youibot initiated a strategic transformation, that is, robots entering the factory.
After 2024, Youibot has become the humanoid robot company that has entered the automotive factory for practical training the most globally, and has achieved strategic cooperation with new energy vehicle brands such as Dongfeng Liuzhou Automobile, Geely Automobile, and BYD.
In the conversation with 36Kr, Tan Min and Jiao Jichao said, In Youibot's plan, the "bipedal humanoid robot" for the home service scenario is the unchanged ultimate goal and also the high ground to maintain "technology priority", and will not be given up. However, before the distant future arrives, the B-end scenario is a more practical route to achieve commercialization.
In Youibot's latest financial report, in the first half of 2024, the AI education intelligent robot and intelligent robot business is still the main revenue driver, with a revenue of 161 million yuan. The income from other industry-customized intelligent robot businesses is 90.879 million yuan, and the consumer-level robot and other hardware equipment achieved an income of 174 million yuan, accounting for 30% of the total revenue.
How much imagination can the factory provide for the humanoid robot industry? The following is the edited conversation between 36Kr and Youibot's CBO Tan Min and Vice President and Dean of the Research Institute Jiao Jichao:
36Kr: At this year's CES, the market wonders why Youibot only exhibited a lawn mower and no humanoid robot products.
Tan Min: The historical mission of Youibot's humanoid robots at the CES Consumer Electronics Show has been completed. Now our mission is in China, focusing on the large-scale application of the B-end intelligent manufacturing scenario. Since 2021, we have placed the global launch of the then latest generation of humanoid robot Walker X at the World Artificial Intelligence Conference (WAIC) in Shanghai, and Walker X has also become the "treasure of the exhibition" of WAIC at that time. Since then, we haven't been to CES, and in the next few years, we will only be in China. In 2024, our first humanoid robot industrial scenario solution was also chosen to be released at the World Robot Conference in Beijing.
Now, the functions of any humanoid robot company displayed at CES, we have actually shown in previous years. Until now, we have not seen any company that can surpass us (the capabilities previously shown).
Since 2023, we have started a strategic shift. Therefore, in 2024, all the company's focus is on the humanoid robots entering the automotive factory, rather than the "showcase".
36Kr: Regarding the strategic focus shift, why choose the car factory among many B-end scenarios?
Jiao Jichao: When Youibot first made humanoid robots, it focused on the home service scenario. At that time, we thought that the technology was already available, but later in the commercialization and testing process, we found that there are many problems that have not been solved, such as hardware cost, weight, and structural design.
Another important point is intelligence. Before the emergence of the large model, humanoid robots could not cope with different scenarios. The home scenario is very complex, ten or a hundred times more complex than the industrial scenario. Due to the different home environments of each person and the different task requirements, these problems cannot be solved by written programs or formulated process task flows.
After in-depth contact with many car companies, we found that the challenge they are currently facing is that the automation rate of car companies is very high, but in the final assembly line, logistics loading and unloading and other links require a large amount of manpower, and these tasks cannot be solved by industrial robots. It requires flexible working space, tool use, and process rhythm requirements.
From the analysis of the scene challenges and production processes, our current product technology can solve these problems.
Taking the handling task as an example, from the perspective of the production process, the handling link is everywhere. If industrial robotic arms are used, one robot may be needed for handling and sorting, and another robot may be needed for placement. The flexibility and generalization of this kind of work is something that industrial robots cannot solve, so humanoid robots are needed to meet this demand.
Tan Min: The discovery and exploration of this scene is not only our one-way exploration. It can be seen that at present, the production capacity of all automotive factories is difficult to increase, because when the market competition is fierce, the labor and labor costs will be rolled up. An important reason why manufacturing customers come to us is that they hope to use humanoid robots to improve efficiency.
The domestic mainstream new energy vehicle factories have mentioned to us that the annual workstation loss rate exceeds 30%, that is, more than 1/3 of the workstations in 10 workstations may need to be replaced once a year, so the pressure faced during recruitment is very high. The labor cost is getting higher and higher, or there is a lack of manpower. The "Manufacturing Talent Development Planning Guide" previously released by the Ministry of Education, the Ministry of Human Resources and Social Security, and the Ministry of Industry and Information Technology predicts that by 2025, the talent demand gap in the ten key areas of China's manufacturing industry will be close to 30 million people.
The labor gap and the loss rate have a very significant impact on the production capacity. But if 10% of these 30 million gaps can be allocated to humanoid robots, it means that this will bring a demand of 3 million units, and the value of the humanoid robot industry will definitely increase.
36Kr: You mentioned that robots can help the factory save some labor costs. How much cost can be saved specifically?
Jiao Jichao: We believe that the cost savings are mainly reflected in solving the problem of "labor shortage". In the significant upgrade process from Made in China to Intelligent Manufacturing in China, the biggest challenge is not technology, but the difficulty in recruiting blue-collar workers. Humanoid robots are a crucial part in alleviating the gap between supply and demand of manufacturing manpower, which is its core value in the manufacturing field.
36Kr: In the factory scenario, what is the necessity of "bipedal"? After all, it occupies a large part of the cost.
Jiao Jichao: We believe that bipedal is the future trend. This is because the usage scenarios and scope of the final solution of the humanoid robot will be generalized, and the usage scenarios of bipedal are the most.
Of course, bipedal is not the only solution. For Youibot, we do not only make bipedal. We have also developed wheeled chassis robots. From the development of bipedal to the development of wheeled chassis is a dimensionality reduction. We have developed an intelligent robot equipped with a wheeled chassis, so it is not a big problem to combine the humanoid upper body with the wheeled chassis below.
After the stable development of the humanoid robot this year, the bipedal and wheeled robots will share a set of software systems, and the hardware will also be reused. Therefore, in the future commercial service scenarios, we will not only use bipedal humanoid robots, but also use humanoid robots with wheeled chassis.
In terms of cost, we expect that the cost difference between bipedal and wheeled chassis may only be about 10% or 20% in the next three years or so.
36Kr: What is the reduction in cost due to?
Jiao Jichao: One is that the material cost of the bipedal humanoid robot itself will gradually decrease, In addition, the joints of the humanoid robot and the structural parts after our mold opening and mass production are the two major sources of cost. With the expansion of the application scale of the humanoid robot and the improvement of production capacity, the hardware cost will accelerate to decrease. Sensors are reused, so the difference between them is not particularly large.
But we have not given up making wheeled ones, but insist on developing products guided by customer needs. If customers have needs, then we can discuss with them the specific solutions for the application scenarios, and can also consider using humanoid robots with wheeled chassis.
Tan Min: We need to solve the problem of advancement, that is, to avoid being eliminated within one or two years or three or four years. Now making wheeled ones is only for economic practicality, and in fact, there is no technical threshold.
36Kr: What is the impact of the application of the large model on the research and development of robots, and where is it specifically reflected? Which aspect of the ability of the robot to be maturely applied is most closely related to the large model?
Jiao Jichao: In the concept of embodied intelligence, including perception, positioning, mapping, decision-making, planning, reasoning, and control, it can be attributed to the manifestation of the AI ability of the humanoid robot or other robots. Among them, the decision-making and planning parts are the technical modules where the large model can play an important role.
From the perspective of the large model and embodied intelligence, it is divided into cognition, perception, and task association operation. From the cognitive perspective, it mainly relies on the common sense and experience of the large model cognition. Therefore, we continue to build the computing power data and technical framework, and we have been accumulating in this aspect in the past few years.
We have gone from the earliest language large model for interaction to the later multi-modal large model for task planning, and then to the current end-to-end large model for dexterous operation.
Internally, we have started various tests and data collections since the second half of last year, including data collection through imitation learning. Based on NVIDIA's Isaac SIM, we have built an internal simulation platform, and all humanoid robots first need to undergo algorithm verification here.
So far, we have accumulated a large amount of data and high-quality labeled data. In addition, in terms of computing power, tens of millions will be invested this year to further expand the GPU cluster.
36Kr: Where is the specific problem with the data and how is it solved?
Jiao Jichao: The bottleneck of the data problem at present is the source. The first-order generated data is useless because the first-order generation requires more information and may make mistakes and have hallucinations.
Some people say that the data trained based on API data alone can be applied. I don't believe it because these data only exist in the virtual world.
First, we need actual data. After obtaining sufficient practical training data in the actual scene, we will gradually improve the simulation scene based on these data. In the future, most of the training will be placed in the simulation scene, but the premise is to have enough practical training data. In addition, in the past two years, we have recruited a large number of standard personnel for data labeling.
At this point, we have three advantages that are different from most companies.
First, we currently have the largest number of factory POC scenarios globally, and our humanoid robots have entered different car factories to collect actual production data. I think compared to some single-scenario companies, Youibot has a faster and broader data source.
The second advantage is that we have our own factory. The third point is in the training in the simulation environment. We are NVIDIA's earliest partner. After we have the first two advantages, a rapid data cycle closed loop will be formed.
36Kr: In the past, the adaptation period for the mechanical arm to be landed on the production line was particularly long. Has it been shortened or lengthened for the humanoid robot now? Can the large model play a role in shortening the adaptation period?
Jiao Jichao: If the large model does not have enough data for training, in a single scenario, most of them are not as capable as the decision-making AI.
But we still solved the deployment problem. For example, for similar tasks, after BYD is completed and then goes to Lynk & Co., it only takes 3 to 5 days to complete the deployment. There is no need for the user to modify the scene, and the boxes can be moved directly for deployment. Although the boxes are somewhat different, as long as they are not completely changed, even if there are changes, we can use a set of data, spend 1 to 2 days for training, and then update it on OTA. And the subsequent operation process does not require engineers, only delivery personnel to follow up.
This set of scenarios is relatively stable, which depends on the fact that we have adopted small AI models in each module. The large model is not omnipotent. At present, there are some situations of abusing the large model, and I strongly oppose this situation.
36Kr: What is the key capability behind the short deployment cycle?
Jiao Jichao: The improvement of AI capabilities and the stability of our self-developed ROSA 2.0 underlying software system. The upper layer of the software system is the algorithm, and the upper layer is the application software. Because we have a comprehensive self-developed ability, we can deeply integrate the software upper-layer application and the algorithm. This system can quickly locate and solve bugs when problems occur.
36Kr: How do you view the business composition of Youibot? Currently, the consumer sector still accounts for a relatively large proportion.
Jiao Jichao: Looking at the whole of last year, the highest revenue proportion is smart logistics, which is also the fastest-growing sector. At present, the revenue proportions of the company's artificial intelligence education, smart logistics, consumer business and other sectors are relatively average, and we are relatively happy to see such a situation.
36Kr: Currently, the strategic focus of the humanoid robot business is on the B-end entering the factory. What is the future strategy for the C-end, commercial end and other scenarios? When is the right time to return to the C-end?
Jiao Jichao: We are not going to return to the C-end, but have been developing towards the C-end. We believe that this time may be around 10 years, but this does not mean that we do not pay attention to the C-end before the arrival of 10 years.
Tan Min: For the B-end, the first is that the industrial humanoid robots in the factory will enter mass production the fastest. After that, the humanoid robots in the commercial service field (hotels, front desks, airports, etc.) will gradually start to be developed and enter mass production and delivery.
In the C-end, the home landing scenario is the most potential application market for humanoid robots. Youibot will take two steps. The first step is the application landing of the companion function. The humanoid robot can understand what the user is doing and accompany them. Currently, we are developing a bionic humanoid robot for the companion scenario; after the companion application is landed, the second step is the landing of the service function. Now the humanoid robot has the ability to understand and decompose complex tasks and finally plan and execute them. Once the motion ability is breakthrough, it will enter this stage.
