Conversation with the founder of the Moon-Mission Embodied AI Community: the robot’s “brain,” the scarcity of talent, and what happened after a hackathon

Original title: 《Interview with the Founder of the Taomoon Physical AI Community for Moonshot Robotic Embodiment Intelligence: The Robot’s “Brain,” the Scarcity of Talent, and What Happens After a Hackathon》
Original author: Dongcha Beating

The organizer of the Taomoon Plan Physical AI Hackathon, the Taomoon Embodied Intelligence Community, focuses on entrepreneurship services in the embodied intelligence track. It brings together researchers, developers, the industrial supply chain, and startup teams, providing services such as events and resource matchmaking for enterprises and science-and-technology innovation teams. What the community wants to build is the close relationship between talent and industrial resources.

At the Taomoon Plan Physical AI Hackathon onsite, some people asked a robotic arm to act as a dealer dealing cards, while others played ice hockey. One team ran to a basketball court and used data collection to record their shooting motions, trying to get the robot to learn how humans jump and shoot.

Embodied intelligence is currently in a rather strange situation. It is placed in the imagination of a market with trillions in scale—valuation, capital, and policy hype are increasing day by day—yet stable delivery and large-scale deployment are still on the way. The models on the screens are already very good at answering questions, but in the physical world, picking up a cup, moving a chair, or dodging a temporarily moved table becomes much less easy.

For decades, text has been accumulating on the internet, but how the body senses, judges, and exerts force still has no ready-made database.

Only after the brain becomes smarter does the body get a chance

Wang Mingyue worked on products and strategy at Hammer Technology and Xiaomi, served as Product Director for the Xiaoai speaker, and also built whole-home smart solutions. By late 2024, she decided to enter the embodied intelligence field. Instead of building a robot, she started from Tsinghua University’s embodied intelligence club and built a community for the entire industry. The Taomoon community has now been operating for more than a year; it has more than fifty thousand members, including more than 300 PhDs in the embodied track. It has organized dozens of in-depth technical communities and carried out more than 50 offline events.

Wang Mingyue’s career has always been at the intersection of software and hardware. The products she previously worked on required inserting invisible software into visible things. After GPT-3.5 came out, she felt that the “brain” suddenly got smarter. Robots were no longer just a combination of mechanical structures and preset motions; they now have the chance to be re-understood.

She does not see this understanding as a contest that will end in five years or ten years. What embodied intelligence needs is a different kind of data—harder to accumulate. Language models can read the text humans leave behind; robots, however, must collide with the real world, make mistakes, and adjust to know how heavy a chair is, whether a cup will slide, and where the body should go to avoid obstacles when they appear.

Dongcha Beating: Please first introduce yourself—why did you move from your past career experience into embodied intelligence?

**Wang Mingyue:**I’m the founder and CEO of the Taomoon Embodied Intelligence Community, and also the initiator of this Taomoon Plan hackathon. I’m an 80s-born, and after working for more than ten years—at Hammer Technology, then later at Xiaomi doing product and strategy—I served as Product Director for the Xiaoai speaker, and also worked on whole-home smart.

I’ve always been doing things that combine software and hardware, leaning toward intelligent terminals. I read the Tsinghua MBA in 2020. By 2024, after GPT-3.5, the brain became smarter. I think the robot body itself will become more intelligent too. I’ve also been thinking about my career pivot—how to connect my past experience to new trends. I think nothing fits me better than robots, because this is the most complex form of software-hardware integration, and it also involves the future division of labor for human beings. So by late 2024, I made a firm decision to come into this industry.

Later I happened to run into Tsinghua’s club mechanism. I applied to the embodied intelligence club. We used that platform to link some top founders and resources. Then we realized that good resources shouldn’t only stay within Tsinghua, so we branched out and created the Taomoon Embodied Intelligence Community, opening it up to the whole industry.

Before this hackathon, we did more closed-door or customized events. Over the past year or so, the community has accumulated PhDs in the embodied direction and people from upstream and downstream of the industry, and we’ve also established connections with many top embodied company founding teams. Taomoon grew out of the Tsinghua system, and this event also received support from relevant Tsinghua institutions and teachers. Tsinghua has produced many technology startups, but there are relatively fewer community-based organizations. We wanted to do this differently.

Dongcha Beating: How big is the embodied intelligence industry in China right now in reality? Does Taomoon gather a batch of people so quickly—it’s more like a small circle, or is it already a mature industry?

**Wang Mingyue:**It’s an industry with very sharp contrasts.

First, its imagination is huge. If embodied intelligence and robots can truly replace a portion of labor, then in principle they could generate as much GDP as the amount of labor it can substitute, and even do some things humans can’t. It’s also connected to aerospace and to the journey from Earth civilization to interstellar civilization.

This is not something that can be completed in five or ten years. With AI and robots, in the future humans may be able to cross from Earth civilization to interstellar civilization. At that time, GDP won’t be something today’s Earth GDP can contain.

Of course, this industry also has bubbles, but they need to be separated and viewed differently.

From revenue and profits, yes, there are bubbles. But in terms of the possible changes it might bring to humanity and the future, you can’t simply call it a bubble. The current technology is too complex and too early-stage. Whether To C or To B, there hasn’t truly appeared large-scale, stable demand yet.

Judging by revenue and profits, it’s still a small industry. Judging by national strategy, influence, and future possibilities, it’s also a big industry.

Dongcha Beating: What’s the difference between a large model made for robots and the language models we encounter in daily life?

**Wang Mingyue:**Embodied large models still don’t have a unified consensus. After years of development, large language models have reached some consensus on technical routes, but embodied models are still each taking their own paths, and the debate is ongoing.

I don’t have a technical background, so I’ll share my understanding after long-term exchanges with technical experts in the industry.

Language models have a natural advantage in data. Over more than 30 years of internet development, all the text and historical records and books left by humans can become training corpora. Previously, multimodal models added sound and vision, but there is still a big gap between that and truly entering the physical world to form a world model.

Embodied intelligence has a body. It must interact with the physical environment, and only through interaction can it learn, judge, and make decisions. Language model data is just the foundation; robots still lack a large amount of interaction data from the physical world. When people pick up cups, move chairs, and avoid obstacles—after they run into hindrances—how the body senses them and what happens next—this kind of data hasn’t been accumulated by systems the way text has.

Autonomous driving can be seen as a very vertical embodied form. Cars only act within the driving scenario, and some data has already been accumulated. But truly general embodied models require multidimensional, multimodal data; collecting that is hard and very expensive.

In the industry, different routes have been tried: data collection, simulation data, and more recently, first-person perspective data collection has been quite hot. Once data accumulates to a certain baseline quantity, an exponential shift may occur.

About a year ago, I asked many people when the embodied intelligence GPT moment would come. Some said ten years, and later it became three to five years. In this hackathon forum, the judgments from founders, researchers, and investors were more optimistic—some believed key progress could be seen in one to three years. The speed is increasing, and it’s not linear.

A large language model is a crucial part of embodied models, and its progress will also drive embodied models. But embodied models naturally need more dimensions of data. In the end, will it be VLA, world models, or something else? There’s still no conclusion.

The scarcest thing is talent

Amid all the excitement in embodied intelligence, the easiest thing to treat as the “real” work is “getting on the ground”—building bodies, brains, models, and a company that can be valued.

But Taomoon wasn’t originally designed as an organization based on a business plan. Instead, it grew out of constantly helping people find co-founders and resources.

A community is something that’s very hard to compute clearly. The boundaries of technical projects are relatively clear; a community’s boundaries follow the people. Some people come looking for engineers; some look for factories; some look for financing. And some just need to meet someone who can understand what they’re doing.

For the community, Wang Mingyue cares less about whether it’s big in size—talent density matters most.

**Dongcha Beating:**Taomoon positions itself as the Y Combinator of the Physical AI field—that’s really interesting. And why didn’t you directly build a startup, but instead started with early-stage investment projects and building a community?

**Wang Mingyue:**In China, communities in the internet era aren’t as popular. One reason is that the dimensions of the internet are relatively few, so alignment on background needs isn’t that strong. But embodied intelligence is much more complicated. For example, in Tsinghua, in our community there are people from different backgrounds—mechanics, automation, materials, chemistry, interdisciplinary fields, economics and management, law, and so on. It requires cross-disciplinary integration; it’s combinational innovation, and many problems can’t be solved by a single discipline alone.

Only by putting people from different backgrounds together can new inspiration arise, and only then can complex problems be solved. This is the necessity for the community.

When I left big tech and worked on the Tsinghua club, I didn’t plan to commercialize it into a community or organization. Later, as I kept doing it, I found that people really needed this community. We used to help people find co-founders and resources purely as a public service; we hardly ever charged money. During that process, I realized that this is something I’m good at—and that I truly enjoy doing.

Embodied intelligence and AI are leading a new era. This era is worth giving birth to new brands. Brands shouldn’t be only tech and product companies; there should also be communities. Taomoon wasn’t planned from the start—it grew naturally.

Some PhDs and some projects I invested in have also invited me to become a co-founder or partner. If I joined a company as a partner, I could bring many people and resources directly. But I already have feelings for Taomoon’s community and this brand. Also, I’m curious: based on instinct and that momentum, how far could I go?

Another reason is that I’ve spent more than ten years doing products, and later also did strategy. If I enter another company to do products and strategy again, it would be a bit repetitive for me. Given my current life stage, I’d rather look at more projects, help them solve resource, people, and bottleneck problems. I have the patience to make Taomoon go far, but I may not have the patience to hold meetings and polish products every day on a single project. My strength is to see what it lacks, and then use Taomoon’s resources to help fill the gaps.

Dongcha Beating: How will Taomoon commercialize?

**Wang Mingyue:**We probably won’t have a problem hitting over one million in revenue this year. There will be revenue from meetings, consulting, and other services. But no matter the valuation or revenue, we won’t be especially aggressive.

What we care more about is talent density. If a person has real substance—strong cognition and professionalism—then even without much preparation, just sitting in the event venue can still generate good content. We’ve been to Silicon Valley and held events during GTC. We’ve invited embodied-direction PhDs from places like Berkeley, MIT, and Stanford, and we also connect with Chinese people working at local embodied companies. We hope to form an international talent network.

Also, we believe early over-commercialization would harm the experience. Truly excellent people don’t lack opportunities. If they feel uncomfortable here, they simply won’t come. Most交流 in Taomoon today is free; bar events may charge a little to cover costs.

In the future, we can do deeper services for commercialization, such as order matching, financing services, and PR services. But for now, we first need to help everyone form a clear sense: at Taomoon, you can make the best friends, meet high-cognition people, and even find co-founders. Investors coming here can also discover good projects. Once the brand and reputation stand firmly, commercialization will naturally come.

In forty-eight hours, who can make machines move

Hackathons are getting hotter and hotter, which also brings in a lot of skepticism. What can be done in 48 hours? Will people just bring half-finished products to demo? Is it just another self-entertaining show?

Hackathons in the embodied intelligence industry are not that simple. In software hackathons, writing websites and calling models—people with creativity can get started quickly. But to connect a model into a system and then let the system coordinate with hardware, the threshold isn’t that low. Wang Mingyue said that among this year’s participants, about 70% have full-stack development experience. The group that can participate in hardware hackathons is very small.

She doesn’t want to describe hackathons as a “cram school” for entrepreneurship. The ranking doesn’t matter. What matters is that through the hackathon, someone gets their first exposure to hardware, first teams up with strangers, and first discovers that they might be able to enter this industry.

Dongcha Beating: The barrier to starting an embodied intelligence company is high. If you hold a hackathon like this, wouldn’t you face difficulties in recruitment and organization?

**Wang Mingyue:**Yes, the barrier for some software-type hackathons is relatively low. If you can use AI tools and have good ideas, you can jump right into Vibe coding. But embodied intelligence is different. In this event, at least 70% of the people have full-stack development experience, understand hardware, software, and models, and know how to connect a model into a system and then coordinate the system with the hardware.

In China, this is a very small group. If you don’t have community-based long-term talent density, then directly holding such a hackathon would definitely be hard to suddenly find that many participants.

But I was also pleasantly surprised by the other 30% who don’t have full-stack development experience. We still need to trust young people’s learning ability. Some contestants hadn’t touched hardware before, but through the hackathon they broke through their own boundaries, began to be interested in hardware, and even wanted to enter this industry later. Sometimes this kind of change can’t be measured in money. When we run an event, we might accidentally change one person’s trajectory and open up a whole new world for them. And a team doesn’t need everyone to be full-stack—having cognition and coordination ability is enough.

We can’t look at penetration rates statically. Today it might be a very small group, and events, outreach, and education can make it grow larger over time. After many funded companies come to find us, the most common question is whether we can recommend some talent. The bottleneck right now is talent. A top university has only so many students, and many people also want to start their own business.

In the future, Taomoon also hopes to go deeper and do things similar to an academy. We will strive for hardware sponsorship so everyone has a fixed place to run experiments, and we can thicken the talent ladder.

Dongcha Beating: What kind of young talent makes you feel they’re excellent?

**Wang Mingyue:**Not just understanding technology—he should have taste, know how to interact with people, and know how to build a company into a more mature state. Beyond technical language, there needs to be maturity in business, people, and organizations that goes beyond age.

If I have a conversation with a young person and the whole rhythm is led by me, I would think: why don’t I invest in myself and do it myself? Truly excellent young people have their own rhythm, their own ideas, and they know when to persist and when to learn.

I invested in an early-stage project. After investing, the project’s valuation grew by dozens of times. The founder gave me the sense that their cognition iterates very fast—if I exaggerate, meeting him in the morning and seeing him again at night, his ideas might already be updated. When I first met him, I still thought I could guide him. Now, maybe it’s the other way around.

When you look at young people, you also need a dynamic perspective. Their cleverness isn’t only in technology—it’s also in financing, team management, and understanding human nature.

**Dongcha Beating:**There have been a lot of criticisms of hackathons from outside. For example, some people think you can’t build a mature product in a short time, and others question contestants bringing half-finished projects to compete. There are even people who say such events have many kinds of chaos and disorder. In this situation, can hackathons still uncover projects worth long-term support—or even investment?

**Wang Mingyue:**I think we need to first adjust everyone’s expectations of a hackathon. It isn’t an entrepreneurship competition, and you shouldn’t expect a team to raise funding for a product they built in 48 hours. Capital and the outside world shouldn’t impose such utilitarian goals on it. It’s of course related to innovation and entrepreneurship, but it’s not the kind of relationship that converts immediately.

Hackathons, just like their name, are first and foremost a kind of spirit. In 48 hours, everyone enters a flow state and creates. Rankings aren’t that important. It’s not the Olympics, and there isn’t a completely unified set of standards. People bring different equipment and have different levels of education and technology. Getting a rank is only because the competition needs fun, and it also rewards truly creative projects—but it shouldn’t be treated like the college entrance exam.

Participants might be a middle school student who needs to move up to higher education, a college student applying to graduate programs, or someone looking for a job. In these 48 hours, they can temporarily forget those identities, focus on a task together with their team, unleash creativity, and try to break boundaries as much as possible. That alone is meaningful. Whether there will be future funding and conversions is a surprise and a gift.

Of course, we will keep looking for excellent teams. Some people might not start a company this year, but next year or the year after. In two or three years, when he and the teammates he met in this hackathon start a company and raise funding together, can you really say it has nothing to do with this hackathon? Some discover through this event that they like hardware and like embodied intelligence and want to keep investing. These seeds don’t necessarily germinate quickly within a short time.

48 hours is definitely too short. The more ideas someone has, the more technically complex the project is, and the harder it is—the final presentation might actually turn out worse. Because there isn’t time to make the demo look great or the PPT look great. When I walked around the venue, I saw some projects that I could tell the teams really liked. In the end they didn’t even make it into the top twenty. I was surprised at the time, and later I figured it out.

It’s our first year doing this. I admit the competition format isn’t perfect. We’ll review it, and next year we might extend the time for the embodied track.

Dongcha Beating: Which projects in this one impressed you?

**Wang Mingyue:**The champion team of the Taomoon Plan hackathon, LoopMaster. They’re from Shanghai Jiao Tong University. Their product, “Massi Cyber Salesperson,” is a self-evolving sales robot. This robot can autonomously iterate sales behavior based on sales targets and a small amount of demonstration, and it reduces sales costs by 40% for supermarkets and small vendors using a hardware sales plus model subscription SaaS model.

Many projects are strong because they’re not just one idea, but because people from different backgrounds can tune and coordinate complex equipment together. Some built a robot dealer dealing cards; some built an ice hockey setup.

There was a team that went to a basketball court, recorded shooting actions using data collection, and then fed that into an embodied large model—hoping the robot or system could understand how people shoot and then debug the system accordingly.

I was also very impressed by a Real to Sim project. Right now, when robots need to solve a specific scenario, they often require engineers to do on-site surveys, which is costly and not efficient. This team used 3D glasses, algorithms, and other methods to record real environments—for example, a factory in Shenzhen—and then fed that back into the robot system. The system first constructs a simulated environment, and then they conduct real operations. People in Beijing might not need to travel to Shenzhen first to coordinate with that factory. Unfortunately, it didn’t make it into the top ten; maybe because the explanation was a bit abstract, but I really liked it.

There was also a “Hugging Robot” project that made it into the top twenty. Students put a cap on a long robotic arm, dressed it, and hugged it. The technical difficulty might not be high, but it contains culture and aesthetics. It’s interesting that students didn’t want to accept a predefined form of the robot.

Filter out the bubbles along with the people

Embodied intelligence must eventually have a body. The body has to enter the scenario, the scenario provides feedback, and that feedback then goes back into the product and model. Wang Mingyue believes that China’s manufacturing capabilities, supply chain capabilities, and density of scenarios give this iteration loop its own speed advantage.

At the same time, the embodied intelligence industry will also face many challenges. Geopolitics, regulation, ethics, competition among peers, valuation bubbles—these will arrive the way they did in the AI industry earlier.

Dongcha Beating: In China, what advantages does embodied intelligence have, and what problems haven’t been fully recognized yet?

**Wang Mingyue:**Embodied intelligence needs to interact with the physical world—it needs a body—and it has to keep trying, failing, and iterating. China’s industrial chains are rich and move very fast. We can quickly build a prototype, find a scenario, enter the market, get positive or negative feedback, and then keep iterating. This speed is something many countries, including the United States, can’t achieve. Manufacturing capability, supply chain capability, and scenario capability are strong moats.

As for disadvantages, I’m not willing to draw conclusions right now. Everyone is still in the infancy stage, constantly trying and failing. Not understanding things yet and lacking talent are normal. The key is whether there’s confidence and whether talent density is thick enough.

That’s also Taomoon’s vision. We hope to put people who want to do this together. It can be “Little Genius” caliber, and we also welcome “Great Genius” with industry experience. They can exchange effectively and team up in the community, and finally build companies. We’ll be alongside them, accompanying and helping. Because it’s still in the development stage, everyone needs each other.

Dongcha Beating: Will geopolitics, policy, and ethics issues from the AI industry gradually spread into embodied intelligence?

**Wang Mingyue:**It will happen for sure, but I’m not too worried. Whatever is supposed to happen will happen. If an industry has no weird things and no risks at all, then it actually indicates it’s not important. The more important an industry is, the more likely it is to have geopolitics, competition, and all kinds of complex problems. There’s nothing new under the sun. When you encounter problems, you solve them.

**Dongcha Beating:**You mentioned earlier that there’s a huge contrast between the industry’s imagination and current revenue. When you zoom into specific companies, the industry is still in its infancy, yet some companies’ valuations are already high. I spoke earlier with an investor who focuses on companion-robot projects; he worries that some products directly facing users still haven’t handled safety, values, and commercialization speed well. What do you think about the relationship between valuations and product maturity?

**Wang Mingyue:**Every industry will have its exceptions. Some companies may seem values-driven and do poorly in products or other aspects, yet still achieve commercial success or high valuations. But exceptions can’t represent everyone.

We still hope to convey a more positive entrepreneurial value system. My view is that price fluctuates around value. A person’s whole life is also like that—you may be overvalued sometimes and undervalued other times. But if you know your weight and measure and understand your value, in the end it will return to rationality.

If the technology isn’t solid and the product isn’t solid—if they haven’t thought deeply about scenarios and commercialization—even if there’s a bubble for a while or it’s very hot for a while, in the end the market and people will forget them. What can remain is still those companies with real strength and accumulation.

We also want to provide positive guidance for newly joined entrepreneurs. Everyone has to truly love what they’re doing. Entrepreneurship comes with many challenges and pain; if you don’t love it, it’s hard to stick with it. For example, me—I truly love what I’m doing now. Like many post-00s, I often don’t sleep until 3 or 4 a.m. If you only calculate economic returns, this bill won’t add up.

At the same time, entrepreneurs also need to do things that are positive for society and for others, so that they get positive feedback and can persist. When facing setbacks for the moment, or sudden waves of praise, be rational and know your weight and measure.

This hackathon turned out hotter than we expected, but our team stayed fairly calm. We did what we were supposed to do—some things were within expectations, and some weren’t done well, which we’ll save for the next improvement. We don’t want to just hold it once, then get hot for a while and be interviewed a few times. We hope to turn it into a brand, a series. For long-term valuable things, we’ll do them first. Whether the bubble is bigger or smaller doesn’t matter—we’ll just let it be.

Go to Taomoon

In 1970, Apollo 13 suffered an accident on the way to the Moon, and the landing mission was canceled. The oxygen in the spacecraft kept running out. The ground control center needed to use plastic bags, tape, and cardboard in the spacecraft to connect a square carbon dioxide filter to a round interface.

This history was later turned into a movie in 1995 by Ron Howard, Apollo 13. In the film, engineers spread those scattered items out on the table and tried them one by one—no one talked about the grand vision of landing on the Moon.

When Wang Mingyue talked about this hackathon, she mentioned one regret at the end. The booth display segment was lively—contestants presented their products at booth after booth in an industrial park—but many judges went to the forum onsite and didn’t get to see that moment.

She said next year they might consider canceling the forum segment, and she wants to take more judges directly to the industrial park.

Original link

Click to find out that BlockBeats is hiring

Welcome to join the official BlockBeats community:

Telegram subscription group: https://t.me/theblockbeats

Telegram discussion group: https://t.me/BlockBeats_App

Twitter official account: https://twitter.com/BlockBeatsAsia