Over the past two years, humanoid robots have dominated industry discussions and stock market trends. Yet, UBTech Robotics—one of the earliest Chinese companies to explore humanoid robotics—has kept a notably low profile.
At the recently concluded Consumer Electronics Show (CES), humanoid robots were everywhere. However, UBTech opted to showcase only a smart lawn mower. This understated presence reflects the company’s strategic shift in recent years. Since 2023, UBTech has been pivoting toward B2B smart manufacturing applications.
According to Michael Tam, UBTech’s chief business officer, the company has already fulfilled its mission of demonstrating humanoid robots at CES. He noted that none of the humanoid robots showcased at this year’s event featured capabilities beyond what UBTech had already exhibited in previous years.
Since going public, UBTech has faced ongoing scrutiny. Like many high-tech companies, it has reported losses, but investor concerns deepened when CEO Zhou Jian and multiple shareholders dissolved their voting agreement, triggering a sharp decline in the company’s stock price. Shares tumbled to HKD 40 (USD 5.1).
Recently, following announcements from several senior executives committing to long-term stock lock-ups, UBTech’s share price rebounded to around HKD 80 (USD 10.3).
UBTech initially set its sights on developing bipedal robots for home applications. However, as commercialization efforts progressed, the company encountered numerous unresolved technical and logistical challenges.
This led to a change in strategy: deploying robots in factories.
By the end of 2024, UBTech had become the company with the highest number of humanoid robots deployed for training in automotive factories worldwide. It has secured partnerships with major automakers, including Dongfeng Liuzhou Motor, Geely Auto, and BYD.
In an interview with 36Kr, Tam and Jiao Jichao, UBTech’s vice president and head of its research institute, elaborated on the company’s evolving vision. While designing bipedal humanoid robots for home use remains UBTech’s long-term goal, B2B industrial applications provide a more immediate and viable path to commercialization.
According to UBTech’s latest financial report, in the first half of 2024, artificial intelligence education and smart robotics were the company’s primary revenue drivers, generating RMB 161 million (USD 22.5 million).
Other industry-specific custom robotics contributed RMB 90.9 million (USD 12.7 million), while consumer-grade robots and other hardware devices accounted for RMB 174 million (USD 24.3 million)—30% of total revenue.
The following interview transcript has been edited and consolidated for brevity and clarity.
36Kr: At this year’s CES, many were surprised that UBTech only exhibited a smart lawn mower and no humanoid robots. Why?
Michael Tam (MT): UBTech has long fulfilled its mission at CES when it comes to humanoid robotics. The company’s focus has since shifted to large-scale B2B smart manufacturing applications in China.
Since 2021, UBTech has moved its product launches away from CES, opting instead for China-based industry conferences. That year, the company unveiled the latest generation of its humanoid robot, Walker X, at the World Artificial Intelligence Conference (WAIC) in Shanghai, where it was the event’s centerpiece attraction. UBTech has not returned to CES since and has no plans to do so in the near future.
At this year’s CES, no company demonstrated humanoid robotics capabilities that surpass what UBTech showcased years ago. The company has yet to see any competitor exceed its previously demonstrated capabilities.
In 2023, UBTech adjusted its strategy, and by 2024, all efforts are now directed toward integrating humanoid robots into automotive factories rather than showcasing them at exhibitions.
36Kr: There are many potential applications for humanoid robots in the B2B space. Why did UBTech choose automotive factories?
Jiao Jichao (JJ): When UBTech first developed humanoid robots, we envisioned them serving home applications. At the time, we believed the technology was ready. However, during real-world testing and commercialization efforts, we encountered several challenges. High hardware costs, excessive weight, and structural design limitations posed significant hurdles. At the same time, before the advent of large AI models, robots lacked the intelligence needed to adapt to the complexities of home environments.
A home environment is far more complex than an industrial setting—possibly ten to a hundred times more complicated. Every household is different, and the tasks required vary widely. These cannot be solved with pre-programmed procedures or predefined workflows.
Through extensive discussions with automotive manufacturers, we identified a key challenge: although automotive production is highly automated, final assembly lines still rely heavily on human labor for tasks such as material handling and logistics.
Unlike industrial robotic arms, which are limited in adaptability and flexibility, humanoid robots offer flexible workspaces, tool usability, and process synchronization.
From both a workflow and technical perspective, our robots can effectively address these pain points.
MT: This isn’t just our own exploration—the industry is actively seeking solutions.
Right now, automotive factories are struggling to increase production capacity because of intense competition and rising labor costs. Manufacturing clients are turning to us to boost efficiency.
A major electric vehicle manufacturer told us that its workstation turnover rate exceeds 30% annually. This means that in a production line of ten stations, more than one-third of positions must be refilled every year—creating significant hiring challenges.
With labor shortages and rising costs, this problem is only growing. A joint report by China’s Ministry of Education, Ministry of Human Resources and Social Security, and Ministry of Industry and Information Technology forecasts that by 2025, the labor shortage in China’s top ten manufacturing industries will reach 30 million workers.
If just 10% of that gap is filled by humanoid robots, that translates to a demand for three million robots—a massive market opportunity.
36Kr: You mentioned that humanoid robots can help factories reduce labor costs. Can you quantify that?
JJ: The real value isn’t just cost savings, but also about addressing labor shortages.
China’s industries are shifting from “Made in China” to focus on smart manufacturing. The biggest challenge isn’t technology. Instead, it’s the growing shortage of skilled blue-collar workers.
Humanoid robots can play a crucial role in addressing this gap.
36Kr: Some argue that bipedal robots are unnecessarily complex and costly. Why does UBTech insist on bipedal designs for industrial use?
JJ: Bipedal movement is the future. The wider the application range, the greater the need for bipedal robots.
That said, we are not limiting ourselves to just bipedal designs. UBTech has also developed wheeled chassis robots, which are much easier to build. In fact, transitioning from bipedal to wheeled robots is a step down in complexity.
We have already developed wheeled humanoid robots, which combine a human-like upper body with a mobile wheeled base. Integrating these two is relatively straightforward.
As our humanoid robots mature, both bipedal and wheeled robots will share a common software platform, and their hardware components will be interchangeable.
In time, both bipedal and wheeled humanoid robots will be used in commercial applications.
Cost-wise, we estimate that in the next three years, the price difference between bipedal and wheeled robots will shrink to just 10–20%.
36Kr: What are the key factors driving down the cost of humanoid robots?
JJ: Two key factors are driving down costs. First, as production scales up, the cost of essential materials—such as robotic joints and structural components—will decline significantly. Second, the shift to large-scale manufacturing will introduce economies of scale, further reducing overall production expenses.
Since sensors and core AI components can be shared across different robot models, the cost difference between bipedal and wheeled humanoid robots won’t be as drastic as many think.
However, we’re not abandoning wheeled designs. Our approach is customer-driven—if clients require wheeled humanoid robots for specific use cases, we will tailor solutions accordingly.
MT: Our goal isn’t just to compete in today’s market but to stay ahead for the next three to five years.
Wheeled robots are built for cost-effectiveness—there are no significant technical barriers to making them. If a company builds only wheeled robots, they risk becoming obsolete within a few years.
Our focus is on developing the most advanced humanoid robots, ensuring that we remain relevant even as technology advances.
36Kr: How do large AI models impact humanoid robot development? What’s their most critical function?
JJ: In the field of embodied intelligence, AI contributes to several key areas. It enables perception, allowing robots to recognize and interpret their environment. It supports localization and mapping, helping them understand spatial relationships and navigate their surroundings. AI also drives decision-making and planning, determining the best course of action in a given scenario, and manages reasoning and control, ensuring precise execution of tasks.
Among these, decision-making and planning have seen the most significant breakthroughs with the advancement of large AI models. The evolution of these models has followed a clear trajectory—starting with language models, which enabled natural interaction between robots and humans. This progressed to multimodal models, allowing robots to process and integrate information from multiple sensory inputs for more complex task planning.
Most recently, AI has reached the stage of end-to-end models, enhancing fine motor control and enabling robots to perform delicate, highly coordinated movements with greater accuracy.
Since late 2023, we have been conducting internal testing and data collection to improve our AI models.
We use Nvidia’s Isaac Sim platform to simulate real-world training before deploying robots in physical environments. So far, we’ve accumulated a massive dataset with highly precise labels.
In 2024, we will invest tens of millions of RMB to expand our GPU computing clusters, reinforcing our position at the forefront of AI training.
36Kr: What are the biggest challenges in training AI for humanoid robots?
JJ: The biggest challenge is data collection.
Synthetic (first-order generated) data is often unreliable—it lacks real-world accuracy and can introduce errors or hallucinations.
API-generated training data alone isn’t sufficient. AI models need real-world data to operate effectively in physical environments.
UBTech has three key advantages in tackling this challenge:
- We have the world’s largest dataset of humanoid robots in factory settings. Our robots are deployed in multiple automotive plants, gathering real-world production data that other companies lack.
- We own our own factories. This allows us to conduct in-house training and optimization.
- We are an early AI partner of Nvidia. This gives us an edge in AI model simulation and development.
With these three advantages, we can continuously refine and expand our dataset, creating a rapid AI learning cycle that most competitors can’t match.
36Kr: Industrial robotic arms used to take months to integrate into production lines. Are humanoid robots faster to deploy? Can AI models reduce deployment time?
JJ: If a large AI model doesn’t have enough data, its performance in a single task will often be worse than a rule-based AI.
However, we’ve solved this issue by building a hybrid AI system. This comprises rule-based AI that handles routine operations, while large models enable adaptive decision-making.
For example, when we deployed a robot in BYD’s factory, we were able to replicate the same solution at Lynk & Co within just 3–5 days, without requiring any major environmental modifications.
Even if there are slight variations in objects, such as different-sized boxes, robots can be retrained on new data within 1–2 days and receive an over-the-air update for seamless deployment.
Additionally, after the initial deployment, the robot’s operations no longer require AI engineers—only general technicians for system monitoring.
36Kr: What are the core capabilities that enable such a short deployment cycle?
JJ: The key factors are advancements in AI and the stability of our in-house ROSA 2.0 software system. Our system is structured in layers—at the foundation is a low-level software framework that ensures stability, above that sits our AI-driven decision-making system, and at the top is the application layer, which handles real-world interactions. Since we develop everything in-house, these layers work together seamlessly.
This deep integration is what enables us to quickly diagnose and resolve issues. If a problem arises, we can immediately pinpoint where it’s happening—whether in the low-level controls, AI logic, or application software—and deploy a fix efficiently. That’s what makes our robots not only adaptable but also highly reliable in industrial environments.
36Kr: Right now, consumer robotics still accounts for a significant share of revenue. How do you view UBTech’s overall business composition?
JJ: Over the past year, our smart logistics segment has generated the highest revenue and is also our fastest-growing business unit. However, we’re seeing a balanced contribution across AI education, smart logistics, and consumer robotics. This balance is something we’re happy about—it reflects a diversified business model rather than an overreliance on any single category.
36Kr: UBTech Robotics is currently prioritizing B2B industrial applications. What is the long-term strategy for commercial and consumer markets? When do you expect to fully enter the consumer space?
JJ: It’s not about returning to the consumer market—we’ve always been working toward it. While we estimate that consumer humanoid robots will take around ten years to become commercially viable, that doesn’t mean we’re putting the space on hold. Our approach is centered on steady progress rather than a sudden market entry.
36Kr: How will UBTech expand from industrial applications into the consumer market?
MT: Our strategy follows a three-step roadmap. The first phase focuses on deploying humanoid robots in industrial settings, where they are already proving their value in factory operations. Once that foundation is established, we will gradually expand into commercial service applications, such as hotels, reception desks, and airports. The final stage will be the introduction of humanoid robots for household use, which we see as the most promising long-term application.
In the consumer space, we are taking a phased approach. The first milestone is companion robots—robots designed to understand user actions and provide meaningful interactions. We are currently developing biomorphic humanoid robots specifically for companionship. Once we establish a strong presence in this area, the next step will be full-service robots capable of executing complex household tasks. At this stage, humanoid robots will already be able to comprehend and break down intricate commands. As their mobility and dexterity improve, they will eventually evolve into fully functional home assistants.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Song Wanxin for 36Kr.