Yu Yinan has often moved toward new technology waves before they fully arrived.
After earning his PhD, Yu did not enter academia, as many of his peers did. Instead, he joined China’s internet sector. In 2015, after Yu Kai asked him, “Shall we go build artificial intelligence chips together?” Yu joined the founding team of Horizon Robotics as employee number one.
In the years that followed, Horizon Robotics grew into one of China’s most closely watched autonomous driving technology companies. It listed on the Hong Kong Stock Exchange in 2024. According to 36Kr, it was the largest Hong Kong IPO that year.
At the end of 2024, Yu left Horizon Robotics and founded Vbot, aiming to bring capabilities developed in autonomous driving over the past decade into consumer-grade robots.
Yu has a streetwise decisiveness that is not always associated with technical founders. In his later years at Horizon Robotics, he managed supply chain, sales, finance, and other functions. After starting his own company, he had to oversee the full business. Decisiveness and discipline became part of the job.
He is now leading Vbot through a series of early milestones. According to 36Kr, Vbot has completed a pre-Series A funding round, raising around RMB 500 million (USD 73.4 million). The round was jointly led by Orient Renaissance Capital, Huatai Zijin Investment, and Fosun RZ Capital, with participation from SAIC-backed Shang Qi Capital and Mornway Capital. Existing investors, including Cathay Capital, Capital Today, Hillhouse Capital, Ince Capital, Baidu Ventures, and Borui Capital, also participated.
Vbot has also begun mass production and delivery. Its first product, the Vbot quadruped robot, officially began delivery on May 8, with the first batch of 500 units rolling off the production line. In addition to bringing quadruped robots into home settings, Vbot has also started R&D for humanoid robots.
Vbot co-founder Zhao Zhelun told 36Kr that the company plans to complete deliveries to more than 1,500 users in May. By June, production capacity is expected to rise to more than 2,500 units, accelerating deliveries to users with reservations.
As production and delivery advance, Vbot is also building offline channels. 36Kr has learned that the company has completed the initial rollout of its offline experience network, including experience stores at Raffles City Changning in Shanghai and Solana in Beijing. Vbot plans to enter more large retail spaces nationwide, including JD Mall and Sam’s Club.
Yu has a clear view of the industry’s timeline. He told 36Kr that China could see home robots with initial practical value in 2027. But he is not unreservedly optimistic. “We believe embodied intelligence has only taken the first step of its long march. The industry is far from converging. Product forms, technical routes, and application scenarios are all still evolving rapidly.”
Vbot has chosen to move quickly and iterate in small steps. Rather than trying to build a “100-point” product from the start, it plans to first launch a “60-point” product and improve it through user feedback.
Its next phase will focus on three areas: robot body R&D for full-size humanoid robots that combine motion and task capabilities; world model R&D for general manipulation and mobility; and the development of an embodied intelligence operating system based on an agentic OS architecture. The goal is to support a broader robotics application ecosystem across homes, commercial spaces, offices, and other environments.
The following transcript has been edited and consolidated for brevity and clarity.
36Kr: What is the current order and delivery situation for your first product, the Vbot quadruped robot?
Zhao Zhelun (ZZ): Since launch, total orders for the Vbot quadruped robot have exceeded 8,000 units. We built a dedicated production line with Huaqin and are still in the production ramp-up stage. About 500 units have rolled off the production line so far. We expect to deliver 1,500 units in May, with monthly capacity rising to 2,500 units in June. At full capacity, monthly output can exceed 4,000 units.
36Kr: What does Vbot’s current customer profile look like? What are the main reasons people buy from it?
ZZ: We found that our user profile overlaps heavily with the previous generation of new energy vehicle (NEV) owners. They are around 30–45 years old, mainly middle-class families with children, and they own Li Auto, Nio, Tesla, Mercedes-Benz, BMW, Audi, and Porsche vehicles.
Their main purchase motivation is family companionship. The second reason is that, as an AI innovation product, it allows children to engage with frontier technology. Users also value its early-adopter appeal and the fact that it can be taken outdoors. A quadruped robot is currently the only embodied intelligence product that can conveniently be brought outside for interaction.
All of our current orders come from China.
36Kr: Vbot focused on home scenarios from the beginning. Compared with mass production and delivery for enterprise customers, what new challenges does the consumer market bring, and how have you addressed them?
ZZ: The biggest challenge is that consumer users have much higher requirements for product quality, reliability, and experience than enterprise users, while the robotics supply chain remains highly immature. We apply automotive sector standards to our consumer-grade quadruped robot and have taken measures to address three major system-level challenges.
First, for the mechanical system, we significantly raised the durability standards for joint motors and the robot body. The goal is for the quadruped robot to remain quiet and coherent after 500–1,000 kilometers of use, with no abnormal noise from loose gears. This required us to redefine component precision and lifespan.
Second, for the energy system, to meet families’ need for long battery life, we completely reconstructed the electrical and electronic architecture. We removed the battery swapping structure used in industrial scenarios and adopted a highly integrated computing platform, freeing up space for a larger battery and achieving five hours of battery life.
Third, for the intelligent system, we integrated an autonomous driving-grade computing platform and sensors into the quadruped robot, making it an autonomous intelligent agent that does not rely on a remote controller.
36Kr: Vbot may be among the first companies to bring embodied intelligence products into real home living spaces and form a scaled user feedback loop. Compared with the traditional approach of building a demo first and then validating scenarios, how will its data flywheel operate?
Yu Yinan (YY): We divide it into two steps. The first is the cold start: first build a sellable product with basic task capabilities. The second is building a self-learning system that learns through practice.
When large numbers of robots operate in real scenarios, they explore at the boundaries of their capabilities. The boundary data near the edge of success and failure is the most valuable. We filter this data on-device, upload it to the cloud for secondary processing, use it to train better new models, and then push them to all robots through OTA updates. This cycle drives continuous capability evolution.
36Kr: How will Vbot define and filter data that is useful for model evolution? How much data is needed for scaling laws to become possible?
YY: We pursue data that is high-quality and highly diverse. High-quality means key data that reflects success or failure. High diversity means data must come from diverse scenarios. I think simple repetitive data, or data that is completely impossible to process, has very little value. There is now a rough industry consensus that perhaps million-hour-level, diverse data is needed to achieve good generalization capability. Simply stacking data volume is meaningless.
36Kr: The quadruped robot seems to be a transitional form factor, bridging today’s consumer products and future humanoid robots. How do you view its role?
YY: We believe a complete embodied intelligent agent needs general interaction, mobility, and manipulation capabilities, but the specific form should be determined by the scenario.
In industrial scenarios, a fixed robotic arm is often enough. Commercial service scenarios place more emphasis on human interaction. The home scenario is the most complex: robots need not only mobility, but also the ability to adapt to stairs, carpets, tables, chairs, and many other real-world environments.
We are starting with the quadruped form because it is more likely to enter homes first, but in the long term, completing more complex household tasks will still require evolution toward humanoid structures.
What we are doing now is essentially building a platform-based robotics operating system and core component ecosystem, then mounting different forms according to the needs of different scenarios. The home is only our first entry point. In the future, we will expand robots into all spaces that serve people.
36Kr: The robotics sector has also been trying to use mature technologies from consumer electronics and automotive supply chains. Have you paid attention to this trend, including whether simulation testing can be used?
ZZ: We have used a lot of simulation testing. For example, before hardware design, we validated algorithms and gait in simulated environments. But I believe simulation cannot fully replace the physical world. We once had an aluminum structural component that reached 1,000 kilograms of strength in simulation, but the actual die-cast component reached only 100 kilograms because of internal air bubbles created during the process. This also revealed that materials, processes, and other engineering details are blind spots for simulation. They must be discovered and solved through repeated prototyping and testing.
36Kr: This year, Silicon Valley and China are both talking about robots entering the home, but before last year, most people thought this scenario would not arrive so quickly. How did this shift happen? When do you think home entry will happen?
YY: The home itself is the largest market scenario. This is also the inevitable result of technology reaching a certain stage, because breakthroughs around the world in world models, whole-body motion control, dexterous manipulation, and other technologies have allowed people to see the possibility.
I believe that around 2027, home robots with real practical value will start to appear. They will no longer merely demonstrate capabilities, but will be able to continuously provide high-quality services in some home scenarios. In the future, the home robot market will expand exponentially as the complexity and chain length of tasks robots can handle increase.
36Kr: There are already quite a few players specializing in home companion robots, but their products seem somewhat homogenized. How will they compete?
YY: I think competitive barriers emerge in phases. In the short term, the core competition is speed: who can complete R&D, mass production, and acquire the first users more quickly. In the long term, the competition is about technical depth, such as patent accumulation, brand reputation, product quality, and the user stickiness built from these, meaning users’ switching costs.
36Kr: Vbot is developing a full-size humanoid robot. Technically, where are the synergies and differences between large humanoid robots and your current quadruped robot?
YY: Humanoid robots and quadruped robots have very high technical synergy. On the hardware side, more than half of the joints, computing platform, batteries, sensors, structural materials, and suppliers can be fully reused.
On the software side, the operating system, electrical and electronic architecture, and communication protocols are almost 100% reusable. On algorithms, the higher-level parts have higher reuse rates, such as world models and vision-language models. The only differentiated part is the lowest-level motion-control layer, which is strongly tied to the specific body form. It is like one intelligence system adapting to different bodies.
36Kr: What key R&D work will Vbot focus on next?
YY: First is robot body R&D for full-size humanoid robots, combining motion and task capabilities. Second is R&D for a hand-foot integrated world model covering general manipulation and mobility capabilities. Finally, we will develop an embodied intelligence operating system based on an agentic OS architecture, giving robots a rich application ecosystem that serves homes, commercial spaces, offices, and other living spaces.
36Kr: What impact do you think the mass production of Tesla’s third-generation humanoid robot will have? Will it create a shock?
YY: I think it will be a super event that defines the industry bellwether, just like Tesla Model 3 entering China back then. It will greatly educate the market, align industry understanding, and tell everyone what the right playbook is. But I do not think it will overturn us. Instead, it will drive the entire industry chain and ecosystem to accelerate, just as Tesla once drove Chinese electric vehicle brands such as Nio, Xpeng, and Li Auto, attracting more capital and talent into this sector.
36Kr: Over the past few years, there have been repeated claims that this industry is in a bubble and will soon cool down. How do you see this?
YY: Claims that the industry is “cooling” have existed for years. But we believe the long march of embodied intelligence has only just taken its first step. The industry is far from convergence.
This year, some robotics companies are going public, which will actually be a major positive for the industry. Just like large model companies going public this year, it will stimulate more capital to pay attention to and invest in the sector, making the whole ecosystem more prosperous, rather than creating so-called capital siphoning.
KrASIA features translated and adapted content that was originally published by 36Kr. This article was written by Qiu Xiaofen for 36Kr.