First World Robot Championship Opens! 127 Brands Compete, Industry "Will Let Humans See Unbelievable Things" Translation of Chinese Article to English

8 August evening, the "2025 World Robot Championship" (hereinafter referred to as the robot championship) opened at the National Speed Skating Rink ("Ice Ribbon"). A robotic competition that gathers 26 events and 487 matches will be held for three days from August 15th to 17th. This is the first global robotics competition with humanoid robots as the main competitors. At the opening ceremony, I saw a welcome performance by "Welcome to the Carbon-Based Life World", where humanoid robots wore flower crowns, dressed up, and walked on the runway like a fashion show. Then, representatives from various teams, including Universe Tree, Star Dynamic, Beijing Humanoid Robot Innovation Center, and Mattel Force, entered the competition. The robot championship is divided into three types of events: competitive events, performance events, and scenario events. The competitive events focus on the stability, adaptability, explosive power, and precision control of humanoid robots; the performance events focus on the real-time coordination and group coordination capabilities of humanoid robots; and the scenario events focus on the fine operation skills and intelligentization levels of humanoid robots in scenario applications. In the past three years, with the development of robotics technology, humanity has made rapid progress from laboratory research to commercial and social scenarios. At the recently closed "2025 World Robot Convention", Xu Youjun, general manager of the State-Local Cooperative Humanoid Robot Innovation Center, said that Morgan Stanley predicted that by 2050, the number of humanoid robots worldwide would exceed 10 billion, with a market value exceeding $50 billion. Although this vision is yet to be realized, Xu emphasized that the humanoid robot industry faces extremely favorable opportunities for dual driving development. Shanghai Fuli Intelligent Technology Co., Ltd.'s founder and CEO, Guo Yihong, believes that the technology curve of robotics has been rising rapidly, and the boundaries of scenario applications are expanding continuously. He emphasizes that commercialization should focus on simple scenarios such as patrolling, shopping guidance, and visitor reception, which not only have practical value but also hone the movement and sensing abilities of humanoid robots. "Can't we make a machine that can be used in various ways?" asked Yang Hongwei, CEO of Beijing Accel Innovations. He believes that commercialization should focus on simple scenarios and gradually increase complexity. "We should not just create a demo or sample product; we need to create something that has actual value." At least until now, the large-scale commercialization of humanoid robots is still facing multiple challenges. Data gaps are one of them. The difficulty lies in the acquisition of high-quality data. The cost of collecting data from various scenarios is high, and the amount of data produced is insufficient. In real-world environments, the costs of data collection are even higher. Another challenge is model transfer across different forms. The differences in degrees of freedom, sensor types, and output dimensions between humanoid robots restrict the effectiveness of model migration. According to Chen Jianyu, founder and CEO of Star Dynamic, their solution is to use "data pyramids" and stage-by-stage training. "In the pre-training phase, we try to use data that is not related to specific forms or is not collected from this actual machine." Then, in the real-machine training phase, we adapt and optimize for similar scenarios. However, Chen also admits that cross-form migration is still a major challenge. "Directly migrating from humanoids (robots) to quadruped robots, or from dog robots to humanoids, may not be possible." Hardware inconsistencies and lack of standardization are another major obstacle. Different manufacturers' designs, component selections, and integration approaches create high costs for adaptation and system complexity. "Today, one humanoid robot might have three different control systems on it, leading to a complex and inefficient system," said Chen Mingxu, head of the Dama Institute's Humanoid Intelligence Platform. The construction of a brain-like big data system is also a focus. Xu Youjun breaks down the problem into "perception limitations", "decision-making gaps", and "generalization bottlenecks". He emphasizes that if an intelligent robot does not have thinking and evolving capabilities, its definition is worth questioning. Xu also emphasizes that the key is to build a perception-cognition-decision-execution closed loop. "Currently, most systems are directly from sensing to execution, but below that, cognition and decision-making are still limited." He also concludes that the brain model is not large enough, the small brain model is too small, and real-time output capabilities are insufficient, meaning that current systems cannot support complex scenario execution.