Humanoid robots reach the "last mile" of deployment; generalization ability may undergo a qualitative change this year

The robot performance at the Spring Festival Gala in 2026 has once again made humanoid robots a hot topic of national interest and has prompted the industry to confront a core question: Can robots only sing and dance? When will they truly be able to enter factories and be applied in practical scenarios? This question has also become a topic of discussion among many experts at the Zhongguancun Forum annual meeting.

“The core showcase of the robot performance at the Spring Festival Gala is the motion control technology. The stage performance is an important test of the robot’s motion performance. While this is a significant part of the industrialization of robots, for robots to truly achieve industrial application, their deep interaction with the physical world to complete various operational tasks is much more challenging,” said Tang Jian, CTO of the Beijing Humanoid Robot Innovation Center. He pointed out that the current critical bottleneck in the development of the industry ultimately lies in insufficient generalization capability. When robots enter unfamiliar environments and face changes in their surroundings or adjustments in operational processes, the success rate of task completion significantly decreases. Tang Jian also provided a clear expectation: “We predict that this year, the generalization capability of robots will undergo a qualitative leap, achieving large-scale application in industrial, quasi-industrial, and simple commercial scenarios.”

Zhu Xiaoxun, director of Siemens China Research Institute, believes that while the robots that shone on the Spring Festival Gala stage demonstrated that motion control capabilities can be quickly enhanced and generalized through virtual simulation, they have not truly addressed the core pain points in the industrial sector. In industrial scenarios, bipedal walking is not a necessity; a two-wheeled structure is sufficient to meet mobility needs, and the core challenge lies in the end effectors. The entry of robots into industrial scenarios cannot be achieved solely through verification in the virtual world; there are numerous intricate mechanical issues that need to be tackled, and the iteration cycle of mechanical problems is not as rapid as that of virtual model training.

Zhang Yufeng, founder of Wujie Dynamics, proposed a critical judgment on industry milestones: “2025 will be the year of mass production for humanoid robots, and this year will definitely be the year of mass production for operational intelligence.”

As the industry moves from the stage to real-world scenarios, there is always a core controversy: should we prioritize developing specialized robots for factories and home scenarios, or focus on general-purpose humanoid robots? “The crux of this issue depends on whether you are driven by scenarios or by technology,” Zhu Xiaoxun pointed out. The core evaluation criterion in industrial scenarios is ROI (Return on Investment). If there are enough single high-value scenarios in factories, such as just tightening screws, specialized robots can create enormous value. From the perspective of return on investment, prioritizing breakthroughs in specialized scenarios is a better choice.

Zhang Yufeng believes that specialized and general-purpose robots are not mutually exclusive options; rather, they represent a phased development path. As long as value can be created, it is a good embodied intelligent product. “In the current stage of the industry, ‘one brain, one form’ is the most suitable path. The core reason is that the current industry data is extremely scarce, and a multi-form layout will only lead to data fragmentation, which is not conducive to refining the general-purpose brain. True embodied intelligence must revolve around humanoid forms, as the data surrounding humanoids is the richest in the industry.”

In Tang Jian’s view, both specialized and general-purpose robots are needed in the industry. The “7:2:1” rule applies to industrial scenarios, meaning that 70% of scenarios can be optimally addressed by existing industrial automation equipment, which do not need to be replaced; the remaining 20% of scenarios represent the opportunity for humanoid robots. Their mobility, adaptability, and flexibility are difficult for traditional industrial robots to achieve. In commercial and home scenarios, humanoid robots will dominate.

Currently, embodied intelligence technology is accelerating its development, becoming the core direction for the deep integration of artificial intelligence and the robotics industry. It is also an important engine for cultivating new productive forces and promoting the deep integration of the digital economy with the real economy.

Yan Ying, a member of the National Committee of the Chinese People’s Political Consultative Conference, vice chairman of the All-China Federation of Industry and Commerce, vice chairman of the Beijing Municipal Committee of the Chinese People’s Political Consultative Conference, and chairman of the Beijing Federation of Industry and Commerce, stated that Beijing will focus on promoting the deep integration of embodied intelligent technology innovation and industrial innovation, accelerating the cultivation of new development tracks. First, we will continue to strengthen technological leadership. We will promote the critical core technology battle action in Beijing, relying on four new R&D institutions in the field of artificial intelligence, collaborating with universities, research institutions, and enterprises in the field of artificial intelligence to accelerate breakthroughs in key core technologies.

Second, we will vigorously promote platform support. We will continue to layout and build key laboratories in the “AI + Robotics” field in Beijing, promote universities to establish embodied intelligence research institutes, strengthen research on common technologies, and connect the links of R&D, data collection, and manufacturing to help enterprises achieve rapid R&D and mass production.

Third, we will continuously optimize the industrial ecosystem. We will jointly create benchmark demonstration scenarios with central state-owned enterprises in Beijing to accelerate the application of embodied robots. We will effectively utilize the multi-billion-dollar industrial investment funds in artificial intelligence and robotics to precisely support startups and hard tech projects with original innovation capabilities. We will continue to increase efforts to cultivate talent in the “AI + Robotics” field, focusing on promoting the deep integration of the innovation chain, industrial chain, capital chain, and talent chain.