Core Content Summary
Zhao Hongyi, the founder of Yufan Intelligence, started with the initial goal of "liberating workers" as a boy from a coal mine. When he started his business in 2014, he wanted to create household robots directly, but due to the high technical complexity, he turned to developing "eyes for robots" (visual recognition) instead. Over the past twelve years, he has focused on practical applications such as construction sites and communities, accumulating expertise in technology, team building, and understanding various use cases. Now, he has returned to the robotics field and launched the world's first heavy-duty quadruped robot, "Lingmao," which can perform a 720° backward somersault continuously, as well as the Manas spatial large model specifically designed for robots. He adheres to a practical approach of "letting robots learn how to work first, and then making them more human-like," avoiding the hype around humanoid robots and emphasizing the importance of real-world scenarios and user needs.
I. A Twelve-Year Detour in Entrepreneurship: A Path That Turns Out to Be a Moat
Zhao Hongyi initially aimed to create household robots that could see, hear, speak, and move, but he quickly realized that each of these functions could be the basis for a separate company. Therefore, he focused on his area of expertise—visual recognition—and delved into practical applications like construction site access control and community security. For example, construction sites require real-name authentication to ensure workers' wages are paid, and his visual technology could identify faces efficiently in harsh environments at a low cost, making it the only viable solution in the market.
These twelve years may have seemed like a waste of talent, but they actually built three key strengths:
1. Technical expertise (visual algorithms and hardware engineering);
2. Deep understanding of use cases (recognizing that different scenarios have different needs, such as energy efficiency for property management rather than high-end parking lots);
3. A mature team (capable of quickly solving hardware and software integration issues).
These strengths became his foundation as he returned to the robotics field—while others were still making mistakes, he already had the necessary technology and experience.
II. The Lingmao Robot: Somersaults as a Proof of Practicality
Lingmao's ability to perform a 720° backward somersault is not just for show; it demonstrates three key capabilities:
1. Advanced motion control: The robot's ability to maintain balance during somersaults indicates that its algorithms can handle complex movements.
2. Robust design: Despite being lightweight, it has sufficient power to carry a load of 20 kilograms (equivalent to several buckets of water).
3. Cost-effectiveness: Similar robots cost 100,000 yuan, while Lingmao might cost only 20,000 to 25,000 yuan (one-fourth to one-fifth of the price).
Lingmao is designed for consumer use and could potentially help with tasks like carrying items, accompanying the elderly, or even delivering parcels in communities (tests have shown it can open doors and operate elevators).
III. The Manas Large Model: Giving Robots a "Brain That Understands Space"
Traditional large models like ChatGPT are good at chatting but lack understanding of the physical dimensions and spatial relationships in the real world. Manas is a "spatial brain" designed specifically for robots:
- Understanding the real world: It knows things like "how tall a table is" and "where the door is," preventing accidents like knocking over cups.
- Real-time control: It can respond quickly when the robot interacts with the physical world (e.g., avoiding obstacles while walking).
Zhao Hongyi insists on developing everything in-house, from software to hardware. He says, "Companies without full-stack capabilities will be eliminated first—AI software evolves rapidly, but if the hardware can't keep up, high-quality products won't be possible." For example, it took them half a year to develop the joint motors for Lingmao, which have one of the highest torque densities in the industry.
IV. Avoiding the Hype Around Humanoid Robots: Focus on Practicality First
Many companies are focusing on humanoid robots, but Zhao Hongyi argues, "I oppose creating humanoid robots just for the sake of it or to chase capital trends." His approach is practical:
- Prioritize real needs: Washing clothes can be done by a washing machine; there's no need for a humanoid robot that is inefficient and expensive.
- Use cases determine technology: The form of a robot should be based on its intended use. For instance, quadruped robots are more stable and energy-efficient on uneven surfaces (such as construction sites and communities) than humanoid robots.
- Focus on functionality: Robots should first be useful before considering their appearance.
V. Data Security and Business Models: Both Trustworthiness and Profitability
1. Data privacy: Their system is used by 120 million people daily for face recognition, but many algorithms are processed offline (without storing data in the cloud), and they have achieved the highest level of security for face information processing (PIA level 2+).
2. Changing business models: They used to sell hardware as a one-time purchase, but now they charge based on the results. For example, if a property management company saves money by using their system, they get a portion of that savings. In the future, they plan to create a "smart entities internet" where users can pay for services like having a robot design a logo, and the robot will complete the task automatically.
Conclusion
Zhao Hongyi's story shows that entrepreneurship is about solving real problems, not chasing trends. His twelve-year journey was actually laying the foundation for the robotics industry today. His practical approach—focusing on use cases and functionality before seeking novelty—may be the key to thriving through industry cycles. After all, the ultimate goal of robots is to be useful, not just attractive in appearance.