Undeterred by the public display of robotic errors, China is installing such centers nationwide. As of the first half of this year, humanoid robot innovation centers — comprehensive hubs for the robotics industry — numbered 22 at the provincial level or above. At least 90 lower-tier data collection centers are also operating, under construction or in the planning stage across the country. The Wuhan center alone generates 24,000 data entries a day to train robots, with the government providing subsidies. In effect, a "Chinese-style accumulation" is under way, with numerous imperfect robots piling up data by using China's vast market as a testing ground.
In a notice released in June, the Chinese government specified that by the end of this year humanoid robots must be equipped with more than 100 high-value-added application scenarios and the capacity for deployment at 10,000 sites. Events such as the robot Olympics and a half marathon, held one after another, ultimately also serve the purpose of accumulating data. State-run Global Times said of the robot Olympics that closed on Aug. 26 that "spectators see the blunders, but researchers obtain valuable data," arguing that "looking ridiculous in the arena is an essential process for robots to grow quickly."
China accounts for 90% of the world's commercial robot data: "Errors must be collected too for self-evolution"
Chinese local governments began racing to build humanoid robot centers in October 2023. At the time, the Ministry of Industry and Information Technology released its first comprehensive road map for fostering the humanoid robot industry, the "Guiding Opinions on the Innovative Development of Humanoid Robots," which specified the strengthening of public technology infrastructure. Just 14 months later, 22 innovation centers had opened across China. Given that in electric vehicles and semiconductors it took two to three years from a policy announcement to the launch of a first center, the pace is unusually fast.
Now in the third year, concerns about overinvestment abound. Critics say the missteps by local governments seen in the battery and solar industries are being repeated in robotics. Most local governments run the centers in partnership with robot makers, purchasing those companies' robots and producing training data. Operating costs are covered by selling the data externally, but purchase demand remains insufficient. Beijing's Shijingshan humanoid robot training center recently terminated its contract with partner firm RealMan, citing data quality that fell short of expectations and low sales revenue.
24,000 training data entries piled up daily
"Even when they fail, the government pays subsidies"
Analysts say China recognizes the contradiction but is accepting it in the interest of speed. The aim is to secure vast troves of data and gain the upper hand in the technology hegemony contest with the United States. Indeed, the axis of the robot competition between the two countries is shifting from hardware to the robot's "brain." Unlike large language models such as ChatGPT, robot AI models cannot simply use internet text and images, and instead must accumulate actual physical interaction data piece by piece in the field. In a report written after a recent visit to a Chinese robotics industry exhibition, Samsung Securities concluded that "the real bottleneck remains in the cerebrum."
Though it is still very early days, the industry expects that once enough data accumulates, a "ChatGPT moment" could also arrive in robotics. Chen Tao, head of Fudan University's deep learning research institute, said the "biggest problem with current robot vision-language-action (VLA) models is that there is plenty of success data but a shortage of failure data," emphasizing that "true self-evolution is possible only when robots correct errors on their own and convert them into experience."
China is already ahead in the race to secure data. According to U.S. data-labeling firm Scale AI, China accounts for about 90% of commercially available robot AI data, and its data production costs are 60% lower than in the United States. The sheer volume of robots it mass-produces is overwhelming, and it also holds an edge in the workforce and labor costs needed to train them. Guo Ping, chairman of Huawei's supervisory board, recently cited computing infrastructure, data and talent as the three pillars of AI competition, assessing that "China lags the United States in computing infrastructure but holds an advantage in data."
