Li Qingdu, Professor of University of Shanghai for Science and Technology, Founder and CEO of DroidUp Robotics

AsianFin -- Li Qingdu, founder and CEO of DroidUp Robotics, stressed that safety and real-world application scenarios are the two pillars that will determine which companies survive—and thrive—over the next five years.

“Many full-sized humanoid robots on the market today still fall short of safety standards,” Li said at the 2025 Inclusion·Bund Conference in Shanghai. “If you don’t solve the scenario and safety challenges first, true mass production is impossible.” His remarks cut through the excitement surrounding humanoid robots, including Elon Musk’s recent claims that Tesla’s Optimus robots could eventually account for 80% of Tesla’s $1.24 trillion valuation.

Founded in 2021, DroidUp Robotics—formerly Shenzhen Xingzhe Robotics Technology Co., Ltd.—has quietly established itself as a technically capable player in China’s emerging humanoid robot market. The company leverages over 20 years of experience in bipedal dynamic walking and cross-modal learning to serve scientific research, commercial services, and home healthcare applications.

DroidUp’s track record includes the 2015 “Xingzhe No.1” robot, which walked over 134 kilometers on just 0.8 kWh of electricity across two days and three nights, breaking a Guinness World Record previously held by Cornell University—a record that still stands today. In 2024, the company completed a 100 million yuan (US$14 million) Series A funding round and released “Walker II,” the world’s first modular humanoid robot based on bionic tendon-driven technology. It later claimed third place at the inaugural Global Humanoid Robot Half Marathon in Beijing Yizhuang in 2025.

This July, DroidUp unveiled “Walker III,” a 1.7-meter-tall, 38-kilogram humanoid robot. The machine is over 20% lighter than industry peers, moves more nimbly, and consumes 50% less energy thanks to tendon-mimicking drive technology and dynamic coupling energy-saving features. Its operational battery life reaches six hours, making it one of the most practical humanoid robots to date.

Li Qingdu emphasized that mass production is already underway. DroidUp’s core R&D remains in Shanghai, while its Zhengzhou subsidiary focuses on scaling production. The company expects to produce several hundred units this year, with 2,000–5,000 units projected for 2026.

On the topic of industry valuations, Li dismissed concerns of a “bubble.” “From a valuation perspective, most humanoid robot companies are only worth 10 to 20 billion yuan,” he said. “We are far from a bubble stage. Humanoid robots will join smartphones and automobiles as drivers of a hundred-billion-yuan market, and only by solving balance, efficiency, intelligence, safety, and cost challenges can mass production and profitability be achieved.”

Looking forward, Li sees the next five years as decisive for industry consolidation and differentiation. “Companies must find real-world applications that genuinely solve customer problems. Every company starts from the same line, and disruption is always possible,” he noted.

As global demand for humanoid robots surges—Merrill Lynch projects shipments will jump from 2,500 units in 2024 to 18,000 units in 2025, with the global population reaching 3 billion by 2060—DroidUp’s careful, safety-first approach may give it an edge. While headlines celebrate Musk and Tesla’s mass production ambitions, Li Qingdu’s perspective underscores that the race is not just about speed or hype—it’s about building humanoid robots that can reliably function in the real world.

For the burgeoning humanoid robotics industry, the lesson is clear: without rigorous attention to safety, stability, and practical deployment scenarios, all the investment capital in the world will not translate into sustainable growth or meaningful returns. DroidUp’s strategy demonstrates that long-term value lies in responsible, methodical development—not just flashy promises.

(Image source: DroidUp Bipedal Humanoid Robot Demonstration)

The transcript of the conversation with Li Qingdu is follows, edited for brevity and clarity:

Q: You previously mentioned that reducing the price of humanoid robots to tens of thousands of yuan is not necessarily a good thing. So, at what price range do you think humanoid robots can reach a balance point?

Li: Back in 2023, when I was talking to many people in the robotics industry, we discussed the issue of price wars. At the time, I said that domestic companies shouldn’t engage in price wars and shouldn’t follow the path of collaborative robotic arms.

Q: But in the end, no one listened, and a price war broke out.

Li: But I also feel that it’s not really the companies’ fault—many people are probably being swept along by capital. So in my view, what China’s robotics sector really needs is a healthy ecosystem.

In fact, I believe the price of humanoid robots should be like that of our laptops or smartphones. Look at the iPhone: every year, when a new model is released, the price stays the same. That’s because they’ve reached a balance point—it ensures enough profit for innovation while also fostering a healthy consumer environment. Once this balance is achieved, I think the robotics industry will have entered a truly mature and healthy phase. So right now, to some extent, the industry isn’t all that healthy yet—maybe we’re still in the process of sifting through the waves.

Q: What price range are you talking about? 99,000 yuan? 200,000 yuan?

Li: I think around 150,000 yuan is a reasonable balance. Elon Musk once mentioned a $30,000 price point, and that’s become an anchor in people’s minds, but that’s just for general consumer-grade humanoid robots.

It’s just like cars: there are high-end, low-end, and specialized models for different scenarios, and the prices vary accordingly.

But for ordinary people—whether it’s for healthcare and elder care, greeting and guiding, or inspection and security—the core issue is whether the value created by the robot meets expectations.

In my opinion, a price of around 150,000 yuan for a robot is relatively healthy. If the price is too low—just like with electric vehicles—you can’t guarantee the quality of the equipment or after-sales service. Someone might buy one, only to find that the company no longer exists in the future.

Q: The embodied intelligence robot sector is very lively right now. Some people think there’s a bubble in the humanoid robot industry, while others believe it’s underestimated. In your view, is there a bubble in the humanoid robot industry?

Li: From a valuation perspective, humanoid robots are still far from bubble territory.

If you do the math, the current revenue scale of the humanoid robot market is still quite limited. In the future, robots are expected to become a market of hundreds of trillions in scale, comparable to that of smartphones (in quantity) and automobiles. From this perspective, companies valued at 10 or 20 billion are nowhere near enough to match the potential of this market; such valuations are still just a bubble.

However, in this process, there is indeed a bubble, as more and more people are flooding into this field. The participants vary in quality—some have average skills and lack true innovation capabilities. Of course, there are also many professionals and highly capable robotics entrepreneurs.

In my view, the core to achieving mass production and profitability for humanoid robots lies in the technology and the product itself. This is a highly technical and challenging endeavor. Only by ensuring that humanoid robots can deliver a satisfactory ROI (return on investment) in application scenarios, and by reducing costs and improving efficiency, can the market resolve the issue of the bubble within the next one to two years. At the same time, humanoid robots will gradually reach a certain scale and level in specific scenarios.

Q: At the 2025 Beijing Yizhuang Humanoid Robot Half Marathon, DroidUp Robotics’ Walker II robot team won third place, and recently DroidUp also unveiled the Walker III at the WAIC exhibition. Are these demonstrations and athletic events part of the commercial development and mass delivery of humanoid robots?

Li: First of all, whether it’s participating in a marathon or showcasing movement to the public, these are ways to validate the reliability and fundamental capabilities of robots. So, in the marathon, we never aimed for rankings; our goal was to complete the entire course without a battery swap.

As a company focused on lightweight robot design, we have faced skepticism from academia, industry, and investors alike regarding this path. However, we have now proven the feasibility of this new technological approach—lightweight, intelligent, and safe. For us, that was the main objective.

In fact, I believe that before humanoid robots can be widely adopted, they must meet five key criteria: balance, efficiency, intelligence, safety, and cost-effectiveness. Among these, safety should be the top priority for the deployment of humanoid robots.

Among them, the humanoid robot marathon serves to validate the robot’s capabilities in movement and endurance. For example, if a humanoid robot can only operate for half an hour or an hour on a single charge, it not only creates a sense of anxiety, but may also hinder the large-scale deployment of robots in the future, potentially becoming a major technical bottleneck. However, if a robot can run a marathon, it means its reliability has reached an acceptable level.

Safety is even more important. The DroidUp robot stands at 1.7 meters tall and is designed to be lightweight, weighing only around 30 kilograms. This ensures a baseline level of safety even in extreme situations. What remains to be addressed are the robot’s ability to perform tasks and the reduction of costs to achieve commercialization. If these two issues are solved, the ROI for humanoid robots can meet expectations.

In summary, the main focuses for current humanoid robots are still safety, cost, and the ability to achieve commercial implementation.

Q: Regarding embodied intelligence, some people say VLA is a “dumb” model, while others claim that world models are not the technical direction for embodied intelligence. Which view do you agree with more?

Li: The VLA model is more suitable for simple, vertical, and single-purpose tasks, mainly because it doesn’t need to generalize across many problems or understand a wide range of knowledge. Tasks like folding clothes or assembly line work fall within the scope that VLA models can handle.

In contrast, world models are akin to humans in their ability to truly understand the world and make predictions about the future.

So, world models represent the future, while VLA is for the present. The two cannot run in parallel for the long term.

For example, many industrial scenarios don’t actually require extensive world modeling, as that would demand higher server capabilities and significantly increase costs. However, VLA can solve computational power issues with just a single GPU or embedded system.

Additionally, data for embodied intelligence is extremely important. From an academic perspective, there is still a long way to go before these models and data can be used to solve real-world problems through training and algorithmic iteration.

Looking ahead, there will be three main directions for collecting embodied intelligence data, forming an “inverted triangle” structure: First, the most valuable data comes from highly skilled professionals, such as artists, scientists, and doctors, which can be used for training; second, simulation and Sim2Real synthetic data, which is the key to reducing costs; and third, real-world robot data, which can help humanoid robots address certain edge cases.

Q: What are the real-world application scenarios for DroidUp’s “Xingzhe II” and “Xingzhe III”? Do you have a mass production and delivery plan?

Li: For DroidUp, at this stage, it’s more important to focus on internal development and scenario validation. We’re taking a steady and methodical approach—next year will be the phase when we achieve mass production at scale.

This year, “Xingzhe II” has already been deployed in multiple scenarios. With the launch of “Xingzhe III” in July, our main focus for the second half of the year will be to validate its application capabilities. “Xingzhe III” is primarily targeted for use in scenarios such as guided tours, inspections, industrial settings, and healthcare. We expect to achieve mass production and delivery of “Xingzhe III” next year, with production capacity seeing a significant increase compared to “Xingzhe II.”

DroidUp’s core R&D is based in Shanghai, while its subsidiary, DroidUp Humanoid Robot (Henan) Co., Ltd., located in Zhengzhou, Henan, is focused on tackling the challenges of mass production for humanoid robots. This year, small-batch production is expected to reach several hundred units, and next year, we anticipate scaling up to around 2,000–5,000 units.

Q: Since the beginning of this year, robots featured in the Spring Festival Gala have sparked a new wave of interest in embodied intelligence and humanoid robots. There’s also talk in the industry about a so-called “first tier” ranking. What’s your view on this?

Li: First of all, I don’t think there’s such a thing as a “first tier” in humanoid robots—everyone is still at the starting line. Compared to the world’s truly large-scale general-purpose robotics companies, both domestic and global players still have a long way to go.

In reality, many investment institutions may not make their involvement public, and even if they are developing their own projects, they might not disclose them. As a result, what we see is often just the tip of the iceberg.

Finally, in my view, the current valuation of robotics companies—whether it’s 10 billion or 20 billion RMB—is just the beginning. To truly achieve large-scale humanoid robot deployment, we’re talking about hundreds of billions, even trillions. So from this perspective, everyone is still at the starting line, and any company could be disrupted at any time. Since there are still many real-world application needs for humanoid robots that have yet to be realized, the industry’s cards haven’t been fully reshuffled (in a “survival of the fittest” sense). In the future, robots will still need to solve many problems in both production and daily life.

Q: If the bubble bursts, how many domestic humanoid robotics companies do you think will remain?

Li: First, I believe that all investments will yield returns, though a company may end up switching to a different track. From a broader perspective, since we’ve raised such a big pool of fish, it doesn’t really matter whether these fish ultimately end up in the lake or not.

Secondly, from a team perspective, companies need to position themselves well. If you find you can’t compete with the leading players, you can pivot to vertical or specialized scenarios, as these areas still require a lot of humanoid robot products. From an industry-wide perspective, the current number—over a hundred companies focused on robot bodies—is still far from enough.

On the AI front, things may seem tight right now, but I estimate that in three to five years, embodied intelligence technology will definitely become oversupplied.

Ultimately, successful deployment of humanoid robots requires a deeper understanding of physical AI and solving real-world problems. From an implementation standpoint, as long as you’ve worked with physical products and supplemented them with AI capabilities, your company will have an advantage in real-world deployment.

Every industry actually has its own niche segments, and we can jump into any of them at any time—there are always opportunities.

For Joyeed, our goal is to usher in a new era of human-machine symbiosis. More practically, we want to start by serving people well in some simple scenarios—such as guiding and patrolling—by doing these two use cases solidly and reliably, then gradually acquiring more skills and going deeper, ultimately serving more professionals in more challenging tasks.