Q: Professor Li, thank you for joining us. You’re now teaching at Harbin Institute of Technology (an institution currently sanctioned by the U.S.), where you also studied in your earlier years. What has been your main research focus over the years? And what topics are you most interested in right now?

A: I’ve been fascinated by operating systems ever since my undergraduate days. Back then, I was drawn to them as some of the most complex and fundamental parts of software. That interest stuck with me—through my master’s, Ph.D., and now in my research and teaching, operating systems have always been at the core of what I do.

Since 1995, when Windows 95 was released, I’ve witnessed how Windows evolved, how Linux gained momentum, and more recently, how China launched its own HarmonyOS.

Today, as we look to the future, robots are becoming key devices at the edge—and they’ll need powerful operating systems to run effectively. From a national perspective, China has long faced the dual challenge of “lacking chips and lacking the soul,” where “chips” refer to hardware and “soul” refers to core software like operating systems.

In the past few years, we’ve seen a growing recognition—across government, industry, and the broader tech community—of how important operating systems really are. That gives me a lot of encouragement. This field has always been close to my heart, and right now, I’m especially focused on how we can build embodied intelligent operating systems and make them a reality.

Q: When it comes to embodied intelligence, China might not be lagging behind. In some areas, we’re even keeping pace—or perhaps pulling ahead—of the West. Are there any comparable projects or competitors internationally?

A: That’s actually a question we get asked a lot—especially by investors. Speaking specifically about intelligent operating systems for robotics, there aren’t many fully public and clearly defined systems like ours, either in China or abroad. Think of how iOS is publicly recognized as a smartphone OS—we haven’t seen a direct equivalent announced yet in the robotics industry. That said, there are definitely some emerging solutions, including those that rely on large models or the fusion of multiple models.

Internationally, one of the key companies we benchmark ourselves against is Figure AI. They’ve made impressive progress, particularly in multi-robot collaboration and scene understanding. Their demo videos show robots moving fluidly and responding effectively to their environment. What we’re trying to do is approach the problem from the system level—combining system architecture with model capabilities—while Figure AI currently seems more focused on models. I suspect they’ll eventually move toward integrating deeper system-level capabilities as well.

In fact, embodied intelligence can be thought of as a triangular relationship: the hardware body, the operating system, and the algorithmic models. You really need all three working together. Ultimately, the goal is to build a unified ecosystem. You might imagine this as a split effort—some teams focus on hardware, while others bring together system and model development. Figure AI is doing a great job on that front and growing rapidly, so there’s definitely a lot we can learn from them.

Q: You mentioned that the “body, algorithm, and system” must work together—a concept that resembles the broader idea of hardware-software integration. Your team includes members from diverse disciplinary backgrounds. How do they collaborate, and how did they join your team?

A: Our team is a mix of university-based researchers and external collaborators. Initially, my focus was on operating systems, but with the rise of AI, it became clear that we needed to integrate talent from the AI field as well.

So now, part of our team works on models—both large and small—including areas like robot control and reinforcement learning. While we can’t all be experts in every field, we have to understand how models function so we can choose the right ones and integrate them effectively with the system. That’s essential for achieving vertical integration.

The university team focuses on systems, made up of my graduate students, PhD candidates, and a few junior faculty members. They’ve been working with me on OS development for years. We’ve also intentionally brought in people with expertise in model-related areas—particularly natural language processing and visual understanding—so we can better interpret and respond to complex environments.

In robotics, hardware integration is absolutely crucial. We need two types of hardware specialists. First, we have the computing hardware group—they need to understand CPUs, GPUs, networks, I/O, and how to connect these to the robot’s physical joints. Second, we have mechanical experts who understand the robot’s degrees of freedom and mechanical structure. While we don’t design the hardware ourselves, we abstract the mechanical capabilities into computational terms so that we can interface efficiently with robot manufacturers.

Overall, our team is made up of four main groups: system specialists, model experts, computing hardware engineers, and mechanical engineers. This interdisciplinary setup allows us to go deep in both intelligent robotics and AI-driven operating system development.

Q: Starting a humanoid robot venture is a long and difficult journey, but it’s worth sticking with. As an entrepreneur, what kind of support and collaborators are needed for such a marathon?

A: Humanoid robots are a typical representation of the AI era. The core of human civilization lies in intelligence, and if robots or other intelligent agents can mature, they might usher society into a completely new phase.

For example, in the future, robots might autonomously manufacture rockets or even explore space—this would be an entirely new world, one that humanity has never envisioned. Of course, such a grand vision requires comprehensive investment. Take mobile phones, for instance. The transformation from Nokia feature phones to smart devices was a systematic project that involved infrastructure, talent, capital, and more. Similarly, humanoid robots involve a more intricate chain of development: the government needs to guide the direction, capital must be patient, tech workers need to keep innovating, and even vocational education needs to be involved. For example, we are encouraging high school students to participate in embodied AI data collection. Just as the internet relies on vast amounts of data, physical models also require more real-world data to support them. Only with coordinated “running partners” across society can this goal be realized.

Q: You’ve mentioned “patience” several times, and in recent years, the concept of “patient capital” has garnered a lot of attention. As an entrepreneur, have you received support from such capital?

A: I am grateful for the support of our investors. We’ve just completed our angel round of funding, and additional investments will follow. There was indeed some debate during the investment decision-making process for the embodied AI operating system project. Some investors questioned the monetization model of the operating system, as Windows primarily generates revenue from Office, Android makes money through its app ecosystem, and HarmonyOS has invested billions but struggles to monetize directly. However, the project was ultimately approved, especially after a senior investor pushed through the opposition, believing it represented the future direction. It is this kind of truly patient capital support that has allowed the company to be founded and developed to where it is today.

It’s widely accepted within the industry that the input-output cycle in the operating system field is highly mismatched. Take Windows and HarmonyOS as examples: the former required thousands of people for development, while the latter’s investment has reached hundreds of billions. The robot operating system faces similar challenges in commercializing, with many investors frankly admitting it’s hard to turn a profit. But the senior investor who insisted on funding our project saw this as the value of “doing the difficult but right thing.” This belief in the long-term value of technology reflects the growing maturity of the capital market and society’s increasing expectations for deep innovation.

Q: We know that the Yangtze River Delta National Technology Innovation Center collaborates with you. Could you briefly describe your collaboration model with them?

A: The Yangtze River Delta National Technology Innovation Center is an important national-level innovation platform. Its core mission is to support the integration of technological achievements with industrial applications through project incubation and the connection of innovation resources. The support from the Yangtze River Delta Innovation Center has been mainly twofold.

The first aspect is financial support. Our project was rigorously selected by the Innovation Center and received direct investment in the form of angel funding. This not only serves as an acknowledgment of the project but also provides crucial support for the early-stage development of the project.
The second aspect is resource networking. The Innovation Center has built an innovative ecosystem connecting resources, research and development platforms, and industrial needs. As a result, the center has linked us with a wealth of clients, industry application scenarios, and provided market promotion services, offering comprehensive support for the company’s growth. This kind of assistance is vital for the technical implementation and marketization of startups.

Editor: LQQ