At the 2026 World Artificial Intelligence Conference, Yao Maoqing, Partner and Senior Vice President at Zhiyuan, President of its Embodied Business Division, and Chairman and CEO of Mifeng Technology, provided a detailed analysis of the primary obstacles preventing physical AI from progressing from technical demonstrations to large-scale, practical applications.
Key Obstacles to Scaling
Yao Maoqing emphasized that for physical intelligence to achieve true scale, it must overcome three major barriers. The first is the data barrier, where real-world interaction data is extremely scarce and costly to obtain. The second is the representation barrier, as a unified physical representation capable of spanning different tasks, scenarios, and embodiments has yet to be established. The third is the feedback loop barrier, where real-world trial and error is prohibitively expensive and slow, making it difficult to scale effectively.
Moving Beyond Demos
He further explained that the critical factor for physical AI to evolve from being "demonstrable" to being truly "functional" does not lie in the capabilities of individual point models. Instead, success hinges on the development of a general embodied intelligence system that is generalizable, optimizable, and capable of continuous evolution.
Comments