By 9 a.m., Wu Zelei arrived at work as usual. After putting down his backpack and turning on his computer, the screen displayed a drone hovering in a virtual wind field, surrounded by turbulent airflow and constant disturbances. His task for the day was to teach this digital drone to "learn" how to steadily lock its posture in chaotic currents.

At 28, Wu Zelei works as an embodied intelligence algorithm engineer at Ningbo Rudong Technology Co., Ltd.. In Ningbo, artificial intelligence trainers have been included in the city's list of scarce job categories. At the corporate level, the demand for related positions is even more direct.

Wu Zelei, busy at work.

Specially Teaching Robots to "Acquire Skills"

In simple terms, Wu Zelei is a teacher dedicated to teaching robots how to "acquire skills." How does he do it? He offers a vivid analogy: "Simply put, it's like building a bridge between the robot's 'brain' and 'cerebellum.'" The "brain" refers to the AI model, responsible for converting external information into instructions the robot can understand, while the "cerebellum" handles execution, driving the mechanical body to translate those instructions into concrete actions. "Our goal is to make robots more precise, efficient, and stable when completing tasks," he says.

Unlike the common perception of "a human wearing motion-capture equipment to demonstrate, then AI breaking it down frame by frame," Wu Zelei's "teaching" primarily relies on an autonomously evolving intelligent training platform. Initially, he and his colleagues build a simulation training ground on the computer, setting up obstacles like sudden gusts, obstructions, and terrain variations, allowing the platform to control virtual robots through repeated challenges. Once preliminary training shows results, the chip containing the platform is transplanted into a real robot for secondary debugging and continuous iteration in the physical environment, until the model is truly "ready for deployment."

"Decomposing human actions only allows a fixed robot model to perform a fixed routine. But the capabilities trained through our method can be transferred to different types of robots, enabling them to learn by analogy and apply knowledge broadly," Wu Zelei explains.

A robot dog undergoing trials in a virtual scenario.

New Profession Faces "Double Advantages"

As embodied intelligence technology accelerates its implementation, the demand for robot "teachers" is visibly heating up. In July this year, the Ministry of Human Resources and Social Security officially designated "embodied intelligence robot application technician" as a new profession. This means that "people like Wu Zelei" now have a clear career path. For example, at Minth Group, Chief Human Resources Officer Xu Hailan reveals that the company's robot-related business surged 400-fold year-on-year, and this year, it still needs around 40 "robot coaches." On recruitment platforms, companies are offering monthly salaries commonly exceeding 10,000 yuan for positions such as "robot algorithm engineer" and "humanoid robot motion control algorithm engineer."

Job growth is not limited to factory floors. At the Gulou Historical and Cultural Block, the NAVIAI humanoid robot, independently developed by the Zhejiang Humanoid Robot Innovation Center, has officially "started work" as a "smart store manager" for retail capsules. Grabbing items, smoothly delivering them, and even waving goodbye at the end—these smooth actions rely on the daily specialized training by application technicians.

A small robot "trained" by Rudong Technology kicks a ball.

Beyond the rising business scale, the current development stage of embodied intelligence itself has made "coaches" an essential need. Wang Chao, head of the marketing department at Rudong Technology, admits that while robots may appear impressive, there is still a long way to go before they are truly integrated into daily life. "Slight changes in lighting, transitions in indoor floor materials, or even a different object in a structured scene can cause a robot to crash on the spot." Dong Zhongqing, Executive President of the Zhejiang University CCIA Ningbo Center (City Knowledge Technology), shares the same view. He believes that for humanoid robots to leave the lab, they must undergo repeated practice in real-world scenarios. As a manufacturing hub, Ningbo boasts rich industrial clusters, providing ample "training grounds" for robot evolution, which will in turn create a batch of new jobs focused on training, debugging, and maintenance.

Regarding career prospects, Wu Zelei is full of hope. The inclusion of his industry as a national new profession is, in his view, a "double advantage": "On one hand, career advancement pathways will be established, making our work more fulfilling. On the other hand, it will attract more young people to join the field, which in turn accelerates the implementation and iteration of embodied intelligence."