At the end of 2023, after a strategy meeting at Horizon Robotics, Yu Yinan, then president of the company’s smart driving division, joined CEO Yu Kai, CTO Huang Chang, vice president Su Jing, and other executives for dinner. Between courses, the conversation turned to the future of robotics.
As its English name suggests, Horizon Robotics has long held ambitions beyond artificial intelligence chips. Robots were always part of the company’s long-term vision.
That evening, however, the discussion led to no clear consensus. Some attendees were “arrivalists,” who believed robots would eventually dominate Earth, much like the Trisolarans in science fiction. Others were “control theorists,” envisioning a future where humans manage fleets of labor robots.
Yu took a different view. He imagined robots not as tools or overlords, but as companions, beings that could coexist with humans as equals.
A year later, guided by that belief, Yu left Horizon Robotics to found Vbot, a startup focused on embodied intelligence and consumer-grade robots.
On December 23, Vbot unveiled a quadruped robot priced at RMB 9,988 (USD 1,398.3) for preorders. Nicknamed BoBo for its oversized head, the four-legged robot was designed around three key user groups:
- For children, it can chat, play outdoors, and record moments from daily life as a companion.
- For older users, it can carry groceries, provide lighting at night, accompany daily walks, and patrol the home autonomously.
- For technology enthusiasts, its expandable backplate allows add-ons such as cargo baskets for camping, remote parcel pickup, or exploration of hard-to-reach environments.
At its core, BoBo is designed for the home. But bringing a robotic dog into domestic spaces requires unusually high standards for stability and safety. BoBo can traverse varied terrain and walk alongside its owner without a remote, while its exterior avoids sharp edges to reduce the risk of injury.
In an industry where hardware startups often pivot midstream, Vbot stands out for its consistency. Anyone who saw the company’s product design drawings in March 2025 would recognize how closely the final product matches the original plan.
Within just over a year, Vbot moved from design to prototyping, mold development, and manufacturing with unusual precision. During a visit to the company’s office in December, Yu was asked whether that discipline came from his B2B background.
He agreed. “Look at Horizon Robotics’ strategy from 2015. Not much has changed after ten years,” Yu said.
Yu spent nine years at Horizon Robotics, where he was part of the founding team behind the Journey series of AI chips and led collaborations with automakers including Changan Automobile and Li Auto.
Still, Yu sees B2B and B2C as fundamentally different worlds. In the consumer market, he believes stability depends on having a sharply defined product vision. During an interview with 36Kr, he pointed to DJI and Insta360 as examples of companies that have successfully identified personalized user needs.
“For any consumer company, the first step in product design is understanding exactly who you’re selling to,” Yu said.
For him, Vbot has been an exercise in applying that principle.
BoBo’s identity as a “companion robotic dog” was shaped by Yu’s own life as a father of two. He wanted a partner for his ten-year-old daughter to explore with, and a playmate for his three-year-old.
In user testing conducted at a Beijing elementary school, 40% of children described BoBo as a “big brother” after interacting with it. To them, the robot was not just a toy, but a peer that offered companionship and reliability.
That sense of companionship, Yu said, depends on more advanced AI, particularly in conversation and cognition.
BoBo relies on cloud-based large language models such as Qwen and MiniMax for language comprehension and reasoning, while perception, decision-making, and motion control are handled locally using in-house models. This hybrid setup reduces training costs while ensuring responsiveness and reliability, even in poor network conditions. It also allows the company to collect valuable motion data for future iterations.
The following transcript has been edited and consolidated for brevity and clarity.
36Kr: You held several hands-on demo events before launch. Why was that important?
Yu Yinan (YY): Watching a video, seeing a live demo, and trying it yourself can present very different success rates. Videos can be edited. Live demos are controlled. But when users try it themselves, the environment is unpredictable. We wanted people to experience the product’s reliability firsthand.
36Kr: Many startups change direction frequently, yet your early designs stayed intact. Why?
YY: Because we built it for ourselves. The surface demand was to offer kids a playmate, but the deeper need was to help parents get kids outside and exploring.
We identified the “demand behind the demand.” My own family reflects that structure, whether elderly parents and young kids, so I took one of the first prototypes home. By designing from real life, our core needs stayed constant, and we weren’t swayed by external feedback.
36Kr: Why choose the four-legged form factor?
YY: We looked at three main consumer robot scenarios: public service, household labor, and home companionship. The first two suit humanoid or arm-equipped robots, which are still long-term projects. Four-legged robots, by contrast, are the most technically mature and commercially viable today.
36Kr: Do you think four-legged robots are ready to be sold as consumer electronics?
YY: Yes. My baseline is clear mobility and recovery, safe remote operation, and usable battery life. BoBo currently lasts five hours.
For children, it’s like parents buying a computer in 1995. BoBo becomes their first touchpoint with the future of robotics.
For seniors, it can follow, carry, and light the way within a few kilometers of daily life.
For adults, its avatar feature acts as a remote explorer: users can send it to inspect hard-to-reach places or, say, wander near a volcano’s rim on their behalf.
36Kr: Not needing a remote control is one of your main selling points. How does BoBo manage to walk naturally beside a person?
YY: Because it perceives, not just reacts. Other robots use LiDAR (light detection and ranging) for simple 3D mapping, treating everything as an obstacle. BoBo uses vast datasets and algorithms to understand what it sees, like how we know a wall isn’t the same as a curtain or pile of leaves.
That understanding, which aids recognizing what something is, is the foundation of autonomy.
36Kr: What’s your greatest realization from interacting with BoBo?
YY: Probably when it played with its shadow. Most AI-driven hardware today is reactive: you give a command, it responds. But I want BoBo to act on its own. That’s what our company name, Vbot, and our flower-shaped logo stand for: vitality.
For a robot, vitality means being an independent entity that doesn’t need constant instructions. I often push the team to make BoBo display self-driven behavior, essentially loops of curiosity and exploration that hint at self-awareness.
36Kr: Beyond curiosity, what other traits define Vbot’s intelligence?
YY: Three key traits:
- Curiosity, by exploring things it finds interesting.
- Survival instinct, by seeking to recharge when low on power.
- Safety, by avoiding danger.
Those are the same qualities parents hope to nurture in their children. Combined, they enable a robot to act autonomously, not just obey orders.
36Kr: How do you balance cloud-based and on-device intelligence?
YY: We use a hybrid setup. The cloud handles large-scale reasoning via models like MiniMax, while everything on the device—perception, decision-making, control—is self-developed.
36Kr: Some embodied intelligence firms collect hand manipulation data, but Vbot focuses on movement and environment interaction. Isn’t that too basic?
YY: Not at all. It’s like human development. You crawl, walk, and jump before learning complex skills. These early experiences form the “pre-training” for intelligence. The motion data we gather builds real-world physical intuition, which is irreplaceable for future evolution.
36Kr: How does your approach to demand differ between B2B and B2C?
YY: B2B is about common needs, forecasting demand for computing power or automotive chips years ahead. B2C is about personalized needs, spotting subtle signals that others miss. I admire DJI and Insta360 because they created entirely new categories by addressing unmet desires.
A consumer brand doesn’t have to please everyone. Satisfying a passionate niche is enough for a startup. Later, you can expand. But if you chase universal demand too early, you’ll sink.
36Kr: What part of Vbot best reflects your own taste?
YY: The fun. We’re building companions, not tools. That comes from both reason and passion. Taste evolves with experience, but love for what you make is irreplaceable.
36Kr: How do you balance present and future demand, like DJI founder Wang Tao does?
YY: Robotics evolves through a “double helix” of supply and demand, but right now it’s still supply-driven. Nothing else can yet walk with you through complex terrain at an accessible price point. So we focused first on perfecting mobility. Once that’s achieved, new applications naturally emerge, like users who send BoBo to patrol their properties.
Electricity, when first discovered, had no clear purpose either, but it eventually powered everything. Mobility will be the same for embodied intelligence.
36Kr: Opening molds for mass production means higher upfront costs. Are you confident you can sell enough units to offset them?
YY: We’re optimistic and are targeting sales in the tens of thousands. That said, cost is not what drives our decisions. Our priorities are clear, in order: user satisfaction, shipment volume, revenue, and profit.
Molding ensures quality and consistency, but pricing has to remain accessible. For now, we are comfortable with minimal, or even negative, margins. The goal is to earn user trust, not short-term returns.
36Kr: Why did Vbot choose to develop its own motors?
YY: Self-developing components isn’t a strategic rule. If no supplier can meet our specs, cost, and reliability needs, we build it ourselves. The production line may use partner factories, but process design must remain in our hands.
I’ll be on-site in January to oversee production myself. Like Elon Musk once sleeping at his factory, I want to experience that firsthand.
36Kr: The embodied intelligence industry often struggles with delivery. What are your biggest supply chain challenges?
YY: The key is finding what I call “angel suppliers.” It’s similar to fundraising. These partners invest time, resources, and trust early on. At that stage, you are effectively trading vision and execution for their confidence.
36Kr: How do you organize your AI team to handle hardware manufacturing?
YY: The AI era has made information more accessible and learning faster, which allows us to stay lean. We operate like a special forces unit, fewer people, higher efficiency.
Each domain, whether joint drives or module structure, has a single accountable owner. Clear ownership eliminates finger-pointing. When issues arise, we always know who is responsible for resolving them.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This article was written by Fu Chong for 36Kr.