From September 16 to 17, the inaugural Aggregated Intelligence Industry Development Conference (2025) was held in Wuhan. Industry leaders and research institute heads from fields such as intelligent robots, low-altitude flight, and the automotive supply chain gathered to discuss the development of the aggregated intelligence industry.

This afternoon, Zhao Zhelun, co-founder of Vbot Dynamics, shared the latest progress on the company's first product, Vbot, at the event, and gave interviews to multiple media outlets including 42-Hz Radio.

He mentioned that the company's first quadruped robot, Vbot, will begin production in October this year. The company is transferring its trial production experience from the automotive industry to Vbot, noting that the trial process may take longer than initially expected. This phase resembles the 'P-stage' vehicle concept in the automotive industry; Vbot Dynamics will conduct tests on these prototype robots and also provide samples to media and internal beta users. Only after completing this testing phase will the robots be sold directly to end consumers. He stated, "The consumer public beta aligns roughly with our original plan, but the time when customers can actually purchase the product will be later."

Vbot Dynamics holds a very optimistic outlook for next year's shipment volume. Zhao Zhelun expects that the quadruped robot Vbot will achieve shipments of 30,000 to 50,000 units next year. He further predicts that the market size for consumer-grade quadruped robots in 2026 will be approximately 70,000 to 80,000 units, with Vbot Dynamics capturing a 60% market share.

The consumer robotics sector remains in a very early stage. Zhao Zhelun described it as "not yet reaching the Ford moment, let alone the Tesla moment." Consequently, there are still significant gaps in the industry; how to navigate paths untraveled by predecessors, and whether there are pitfalls ahead, remain unknown variables.

Following this in-depth conversation with Zhao Zhelun today, we gained a clearer understanding of Vbot Dynamics' approaches to product definition, R&D, and sales strategy.

Why Consumers Should Buy

Zhao Zhelun, co-founder of Vbot, first discussed the product definition of Vbot: a quadruped robot capable of moving within living spaces and communicating via natural language. Combining these three elements results in a robotic product that requires no remote control.

Why choose a quadruped design? Because over the past few years, grippers, quadrupeds, and autonomous mobility have entered the engineering phase, while bipedal robots and dexterous hands still have some distance to go.

What are the actual user scenarios? Vbot's definition is not limited to factories or homes but is divided into the natural world and the human-built world. Today, robots are transitioning humans from the era of graphical interaction to the era of natural interaction.

In terms of product design, it achieves natural language communication while striking a balance between failure costs and product complexity.

Currently, ordinary consumers show considerable interest in robots, but Zhao Zhelun believes that converting this interest into purchases is a difficult process.

He believes there will be three stages from technology to users. The initial stage is purely technological, featuring many demo videos and academic papers; the second stage involves some productization with improved reliability, yet its utility for users remains unclear; by the third stage of true commercialization, it must clearly articulate what it does while achieving a proper match between price and value.

"So today, in fact, no robot company has truly reached the third stage of properly matching utility with price and value," Zhao Zhelun said. "I believe that what the industry needs right now isn't to find an iPhone; rather, it should first find a Walkman, whose emergence is better suited to the background of that era, along with its corresponding technological innovations and user demands."

At the World Robot Conference held last month, Vita Robotics participated. Zhao Zhelun commented that the exhibition provided them with very positive feedback and could be considered the most 'C-level' booth at the event.

Through communication with users at this exhibition, Vbot Dynamics found that natural language interaction for quadruped robots is equally important, so it has increased investment in this area to better understand what humans say. Meanwhile, Vbot Dynamics hopes to work with partners to make the Vbot an affordable robot product for more families.

Offline stores to be launched by year-end

How to sell the Vbot, which is about to go on sale, is also a core issue. Currently, most robot products are sold online, leading to very high return rates. So how will Weita Power address these issues? Will they directly engage in price wars?

Zhao Zherun analyzes that in the early stages of robot product sales, the industry will not engage in price wars; Vitas Dynamics will place greater emphasis on NPS (Net Promoter Score). He recalls his experience working at Tesla years ago: "At that time, Tesla's cars were not cheap, and the volume was not as large as it is today, with only four thousand to five thousand units sold per quarter, but the NPS was exceptionally high."

In the early stages of the robotics industry, achieving a higher Net Promoter Score (NPS) may well be the key to success.

Meanwhile, Vieta Dynamics has begun laying out its offline experience stores starting from the end of this year, with the first phase enabling customers in major Chinese cities to experience the robot products.

Zhao Zhelun analyzed that while consumers might not need to visit physical stores to purchase 3C products priced at 1,000–2,000 yuan, they would likely want to try drones priced around 10,000 yuan in person. The same applies to automobiles, where it is rare for consumers to pick up a car without seeing or test-driving it first.

Therefore, Vbot can be seen as a product positioned between drones and automobiles. While online channels may generate interest ('planting grass'), consumers still require offline experiences to make purchasing decisions.

Vieta Dynamics' offline stores will integrate multiple capabilities including experience, sales, and after-sales service, adopting a model combining direct operations with partner co-construction to build core experience stores alongside professional distribution teams.

Additionally, Zhao Zhelun believes that robots naturally attract attention, so the company will also collaborate with other offline stores to increase product exposure.

Robot Manufacturing Hasn't Reached the Ford Moment Yet

When discussing the development trajectory of the robotics industry, Zhao Zhelun believes that China's robot supply chain remains insufficiently mature. Robot manufacturing has not reached the 'Ford moment' at all, let alone the 'Tesla moment'.

He analyzes that robots were previously used in B2B application scenarios, with small production volumes for each company. Consequently, intermediate stages had to maintain high gross margins, ultimately leading to very expensive robot components. However, due to low volumes, even with high margins, the entire industry struggled to make significant profits. This is also why leading companies in the sector are increasingly developing their own robot components.

Therefore, Flexiv aims to bring suppliers' gross margins down to a relatively reasonable range as robot shipment volumes gradually increase.

In discussions after the interview, Zhao Zhelun noted that motors currently account for the largest portion of Vbot's Bill of Materials (BOM) cost. Additionally, Vbot utilizes chips with higher computing power and employs LiDAR; these two components are also relatively expensive, whereas battery costs are not particularly high.

He stated that today, the only module in the robotics industry that cannot be shared with automobiles is the motor, because robot joint motors are smaller.

Currently, purchasing one robot joint motor costs approximately 400 RMB. For a quadruped robot requiring 12 such motors, this results in a BOM cost of 5,000 RMB, which is extremely 'terrifying' for small quadruped robots. Zhao Zhelun believes that in the long term, the price of robot motors is expected to drop below 200 RMB.

Below is the transcript of our conversation with Zhao Zhelun, co-founder of Flexiv, slightly edited.

Q: From a practical standpoint, are there still not many consumers willing to pay for robots?

A: Over the next three to five years, I believe this will indeed be a difficult process because there are actually three stages from technology to users. The first is that today it is purely technology; you see many demo videos and academic papers. The second involves some productization—you can actually buy it, its reliability is relatively acceptable, but it’s unclear what use it has for users. The third is true commercialization, where it needs to make a good match between its utility and its value. So, currently, no company has truly reached the third stage.

We hold a viewpoint: I believe that rather than finding an 'iPhone' in the industry today, we should first find a 'Walkman'. The birth of the Walkman was far more adapted to the background of that era, as well as its corresponding technological innovation and user needs, than the iPhone was.

Therefore, we believe that under a robot's technical architecture, the core capability is spatial mobility. Essentially, all of a robot's capabilities are forms of spatial movement, whether legs, wheels, dexterous hands, or robotic arms. The key is how to effectively apply spatial mobility technology in vertical scenarios to help users solve needs they previously could not address or afford. This is what we consider the core issue. Today, our company sees such an opportunity with quadruped robots; they can follow you anywhere and walk any path without requiring you to control them.

Q: What is the core driving force at the system level in the quadruped robot field? Does Vbot have any layout in this area?

A: Why is an operating system needed? The essence of an operating system is that when your computing device already has excellent capability layers, these layers can be cross-organized into various applications. At this point, you need a relatively more universal form. Through an operating system, developers can create applications with less effort.

Therefore, our Vbot features a universal backplane, which we view as a common platform. On this platform, I can mount cameras, robotic arms, and even other interesting components.

Once I possess strong general-purpose mobility capabilities and carry various different hardware on my back, it can form applications. I hope that my general-purpose mobility capabilities can be utilized by developers.

If developers can casually invoke my general mobility capabilities, it could become a mobile refrigerator, a mobile projector, a mobile robotic arm, or a mobile photographer. At that point, it truly becomes an operating system. This is something we believe must be achieved within the next 3 to 5 years.

Q: What feedback have you received for Vbot? How is the product launch progressing?

A: First of all, WRC (World Robot Conference) has indeed given us very positive feedback. One outstanding aspect is that I believe our booth was the most 'C-end' focused in the entire venue. For the following two days, families came with their children and spouses to visit our exhibit; many people visited our booth and interacted with the robot.

Therefore, we validated our initial hypothesis: the fun and utility of natural language interaction are far from being fully exploited. Consequently, we have increased our investment in developing products capable of understanding the language of various family members, including children.

Another point is that due to the strong popularity we received, we further collaborated with our supply chain partners. We aim to make the price as reasonable as possible. Of course, because the industrial chain is not yet highly developed, the initial price may be relatively high. However, we believe in making the product affordable for as many families as possible.

The third point is that we have more meticulously reviewed the entire production process to ensure higher quality. We will begin mass production next month, in October. However, we will leverage our experience from the automotive industry to intensify the trial production phase of our assembly line.

Our trial production process may take longer than originally anticipated. We will conduct multiple rounds of trials, adopting the concept of 'Phase P' vehicles from the automotive industry—using these units for our own testing, including internal media tests and public user beta testing. We plan to extend this phase slightly to allow more people to test and use the product. After gathering feedback, we will then officially sell to C-end consumers.

Therefore, while the time for consumers to participate in the public beta test might remain similar to our original expectations, the time when they can actually purchase the product will be slightly later. Because we want to conduct more testing, we believe that ensuring high quality is essential to making this product even better.

Q: For a robot product like Walkman, do you think the most core aspect is language interaction?

A: There are several points. First, we believe it must achieve excellence in foundational capabilities, including natural language interaction and autonomous mobility. Furthermore, its autonomous mobility utilizes the VLA (Vision-Language-Action) model from the field of autonomous driving.

Second, as seen today with vehicles like the Xiaomi YU7, there is off-cabin language interaction; you can wake it up from outside to open the front trunk. When you wake it up, its lights flash, indicating a wake-up state. This is very similar to robots. To be honest, the application space inside a car isn't that large, but imagine if you could wake up this robot and say 'come here' or 'go over there'; this provides flexibility.

Third, as seen on the XPeng P7, when you wake it up from the passenger seat, it moves towards you; similarly, waking it up from the driver's seat causes it to move towards that side. This is essentially the same as how we handle the robot's head. It primarily achieves head positioning through sound localization.

The third point is about the VLA models discussed extensively today. We believe they are more necessary in living spaces. Some might feel they are somewhat optional in traffic environments because voice control is too structured. In a car, operations are limited to lane changes, turning left, turning right, and U-turns. However, when communicating with a robot, what you need to say is entirely described by language, and you may not even be able to describe it clearly. At this point, we believe the 'Language' component becomes significantly more important.

Q: What key breakthroughs in the past two years have driven the product towards mass production, and what is your judgment on future shipment volumes after mass production?

A: Regarding robots, I feel we are standing on the shoulders of giants. Why did we choose the quadruped form? As mentioned earlier, a quadruped is the most stable morphology for a robot body. The quadruped has gone through at least two waves, and we might be the third. The first wave was Boston Dynamics' Spot, which truly打通ed (connected/realized) all-terrain mobility capabilities. Five years ago, everyone saw various videos of Spot demonstrating better mobility than humans.

The second wave involved Unitree, which made significant breakthroughs by effectively leveraging China's supply chain to make the product quality and architectural design more mature. Many overseas research institutions favor Unitree's Go2 because it is much cheaper than Spot while maintaining good quality, further maturing the robot body. For us, we aim to further optimize the product structure and quality of the body. Additionally, given the two major opportunities mentioned earlier—one being edge-side AI capabilities—how can we apply algorithms accumulated from autonomous driving to robots?

The third aspect is the application of large language models to enable natural interaction on robots. One line represents the main trajectory for robotics, while two dotted lines originate from autonomous driving and large models. The convergence point of these three lines is the robot we aim to develop.

In terms of shipment volume, we estimate that next year will be our first full calendar year of sales, with shipments reaching 30,000 to 50,000 units.

Q: Are there plans for self-developed joints and chips?

A: Today, China's supply chain is highly mature, which has made the possibility of building robots feasible. However, the robotic supply chain itself remains immature, resulting in high robot prices and a lack of standardized production processes. If you visit current robot manufacturing factories, you will see their automation levels are far lower than those in the automotive industry; assembly is largely manual. Therefore, we believe there is significant room for improvement across the entire supply chain.

Since this industry is primarily B2B, each company’s production capacity is small. Every link in the middle demands a 50% gross margin, making components expensive. Despite the high prices, profits remain low due to the small volumes.

We hope to drive up volumes to bring the gross margins at each component stage down to a relatively reasonable range. To achieve this, we recently realized why Tesla and BYD initially chose to self-develop so many components. Partners cannot truly believe you will reach such volumes or offer costs low enough to make the business viable.

For example, joint motors currently account for a significant portion—about one-third—of a robot's cost. Motor manufacturers need to make money too, so we have already begun self-developing our own motors. For other components, we will collaborate extensively with existing automotive partners. There is still vast potential in our supply chain because volumes have not yet taken off. We are nowhere near the 'Ford moment' for robots, let alone the 'Tesla moment'.

Secondly, I see immense hope in today's robotics, sharing similarities and differences with the electric vehicles I observed ten years ago. A decade ago, I was working on smart electric vehicles; I am deeply passionate about this sector, which has developed rapidly. Automotive culture originates from Europe and the US. Although many foreigners come to Chinese auto shows annually to learn, they are primarily looking for technologies they can utilize. Fundamentally, cars remain an European-American industry. If we aim to create a global automotive brand, it will be very difficult, as it is unlikely we can completely occupy others' strongholds or fully dominate their markets. Establishing automobiles as a 'China Native' industry is extremely challenging.

But today, the automotive industry has driven an improvement in China's entire supply chain. This includes all electrification components like batteries and motors, as well as the sensors, controllers, and chips for intelligence. In my view, these can be completely reused in robotics. Currently, China holds an absolute advantage in this sector globally; we will absolutely make robotics a 'China Native' industry. It is very core that Tesla cannot build robots without China's supply chain.

So, if we return to this entrepreneurial journey, I realize it is basically a 20-to-30-year endeavor: to create robots made in China and distribute them globally. At least, I want them to become as popular worldwide as European and American automobiles were in the previous era, or Japanese and Korean electronics in their respective eras. This is what I see and why I am so determined to embark on this startup.

Q: From perception to decision-making, what do you think are the core challenges currently faced?

A: There is indeed some industry consensus on this. Compared to AI, robotics severely lacks data. Today's data comes from internet text and images; there is no data regarding robot movements or dynamic states. The only truly genuine form of Embodied AI and the only fully validated pipeline so far is autonomous driving. Furthermore, autonomous driving relies heavily on first-person perspective data from drivers. In fact, many companies today are replicating this chain by collecting their own data, establishing large-scale data acquisition centers, including Tesla and several domestic companies. While self-collecting data sounds relatively limited, in terms of data volume, this is a standard industry problem: without data, subsequent AI development is impossible.

There are three stages in the data aspect. The first stage is a 'start-up' phase. We are currently working on models for autonomous mobility in living spaces. In fact, similar to how all autonomous vehicles operate today, we have collected a vast amount of data on autonomous mobility in living spaces. If media teachers visit our company in Beijing, you will likely see dogs downstairs everywhere, conducting massive data collection. This is a necessary starting phase.

The second stage explains why we truly want to deliver products to users. It must go from product to user, allowing more users—tens of thousands, hundreds of thousands, or even millions of users—to co-create data with us and provide more scenarios. Data collected solely by ourselves is always limited; generally, our engineers find it difficult to enter residential communities, so the scope remains restricted.

The third stage is when we believe robotics may bring new prosperity. Today, discussions focus more on generated data and simulation data. This capability is less critical in autonomous driving, where real closed-loop data accounts for 95%. However, in robotics, we believe that simulation and generation will bring greater prosperity, but this is predicated on having a solid foundation of real data. We believe the third stage will bring about generation and simulation.

So, taking the first step and looking at ourselves, we will pragmatically start with data collection.

Q: Besides technology, is talent the other biggest challenge facing the industry today?

A: This issue actually occurred in the autonomous driving sector about four to five years ago. In my personal experience, five years ago, including teams expanding autonomous driving capabilities at companies like Li Auto, there was a significant effort to recruit industry talent.

At that time, it was discovered that there was a severe shortage of autonomous driving talent across China. Subsequently, in 2022, 2023, and 2024, a large number of autonomous driving talents were cultivated through campus recruitment and training programs. Today, looking at the transition from autonomous driving to robotics, China remains an excellent fertile ground for talent, including those returning from overseas and students from Chinese universities, whose quality is exceptionally high.

Since robotics is a completely new industry, even those who have worked on robots for ten years may not necessarily excel in today's Embodied AI. Therefore, I believe it is more important to integrate more young people into this industry to work on new projects and challenges, which will bring a substantial amount of practical, hands-on talent to the field.

The industry currently lacks less theory; talent cultivation must focus on practical application to get projects done. Although this is indeed a short-term challenge—finding someone with direct relevant experience is difficult now—I expect that in two or three years, China will have a large pool of such talent who have completed several projects.

Q: Will Vbot use product pricing advantages to achieve differentiated competition upon its launch?

A: The industry has not yet reached a stage where pricing advantage determines product success. The most critical factor today is not whether the price is high or low, but whether the product offers unique value.

We are still determining the internal pricing, as it depends on our overall cost considerations. We will certainly offer a competitive price, but at this stage, competing solely on price is not the goal. Internally, we refer to NPS (Net Promoter Score); what matters most is whether users find the product useful enough to recommend it to friends after purchase.

Because I worked at Tesla very early on, I have a deep understanding that back then, Tesla cars were actually quite expensive. During the Model S and Model X era, to be honest, the volume wasn't large. What left a strong impression on me was that nationally, sales per quarter were only around 4,000 to 5,000 units, yet the Net Promoter Score (NPS) was exceptionally high. However, Tesla's core strategy was user referrals: when one Tesla owner referred three other owners, they would receive a set of wheels as a reward. Owners were enthusiastic about making these referrals because it essentially brought a new experience, which remains the core of this industry today.

Q: Will we open offline stores so that users can try out the products first, thereby reducing the return rate?

A: Since the WRC, we have been discussing for nearly two months how to sell robots. If you categorize products for sale today, there are several types. For a 3C product priced between 1,000 and 2,000, you can likely buy it online without going offline. However, if you want to buy a larger digital product, such as a drone costing close to 10,000, you might need to visit a physical store and chat with the staff. The most extreme case is cars; it is very rare to pick up a car without test-driving it first.

So today, robots essentially sit somewhere between drones and cars in our view, making offline experience very important. It’s likely that online users are already placing small or large deposits, meaning they’ve become quite 'interested' in us. But to truly confirm their commitment to the product, they still need to try it out offline.

We will indeed begin setting up offline experience centers by the end of the year, allowing online-interested consumers—at least in the first phase within major Chinese cities—to experience and purchase our robots. This is what we will focus on in the next stage.

Sales are extremely challenging. When we look back at robots, truly no one had ever sold them before. Today, there are people selling robots, but in my view, they are doing it very crudely. Selling them like 3C digital products won't work either; and selling them like cars isn't viable, as that approach is too costly. Selling them like 3C digital products lacks the necessary experience, so neither approach works.

I’ll share a few points we’re currently discussing internally, which I think are quite insightful. First, we’ve realized that robots themselves generate traffic. Many stores today rent out robots on weekends to perform actions and attract customers. Therefore, we believe there’s a significant opportunity for our robots to partner extensively with existing offline stores: not only to help them drive foot traffic, but also to expand our own brand visibility. This is an opportunity we’ve identified, because we simply don’t have the capital to open numerous stores ourselves. This approach primarily serves as a means of exposure.

Secondly, regarding the user experience, we believe that the robot's experience is very important. We will open offline experience stores, and I believe these offline experience stores should fully handle the experience, sales, and even a certain level of after-sales service. This way, consumers will feel more confident when making a purchase. If the product turns out to be useless after buying it offline, they can always come back to ask you questions or report any issues, which is quite important.

This part would be too heavy if we did it entirely on our own. Therefore, we will adopt a model combining direct operations with joint development by partners, collaborating with professional distribution teams to establish core experience stores. Our goal is to integrate sales, product experience, and after-sales service into a unified system. This ensures that in the first phase, customers in major cities like Beijing, Shanghai, Guangzhou, and Shenzhen can experience the products firsthand. After deciding to purchase on-site, they can return to us for any issues, which is the core of our strategy.

Thus, from our perspective, there are two key aspects: top-level exposure, where online presence will definitely be prioritized with extensive packaging, while offline channels remain essential. Additionally, at the experience layer, we aim to innovate to help users genuinely perceive the product as purchasable. Both layers will involve innovative approaches.