On July 19, 2026, at a forum associated with the World Artificial Intelligence Conference, Yao Maoqing, who serves as a Partner and Senior Vice President at Zhiyuan, President of the Embodied Business Department, and Chairman & CEO of Mifeng Technology, emphasized that in order to achieve scalability in physical intelligence, three critical barriers must be overcome.

The first barrier is the data barrier, characterized by the scarcity and high cost of acquiring real interaction data. The second is the representation barrier, where a unified physical representation that spans various tasks, scenarios, and agents has yet to be established. The third is the closed-loop barrier, where real-world trial-and-error processes are expensive, feedback is sluggish, and scalability proves challenging.

Yao Maoqing further pointed out that the crux of transitioning physical AI from mere demonstrations to practical, real-world applications lies in constructing a general embodied intelligent system. Such a system should possess the capabilities to generalize across different contexts, optimize its performance, and evolve over time.