【Focus on the 2026 Zhongguancun Forum Annual Conference】 No data feeding, no training—equip the robot with a "brain" directly

（Source: Workers’ Daily）

While the global artificial intelligence race is still competing over computing power scale and data volume, a completely different technical approach is being quietly advanced in China. If we had to explain its difference in one sentence, it is this: instead of feeding robots data, you give them a “brain”—so they can learn to recognize the world on their own, like humans.

The practitioner of this exploration is Tongren Intelligence’s Chairman, Wu Yiming, a researcher at the Xi’an Institute of Optics and Precision Mechanics of the Chinese Academy of Sciences and a doctoral supervisor. At the 2026 Zhongguancun Forum annual conference, during the parallel session on hard-tech investment and development, Wu Yiming delivered a speech titled “Embodied Intelligence Is a New Paradigm for Intelligence Science.”

“Today’s large language models’ understanding of the world comes from human textual descriptions of the world, not from the world itself.” Wu Yiming said in an interview with a reporter from the Workers' Daily. What people call “training” is, in essence, statistics—finding answers through probability in massive databases. The machine does not truly understand the world.

Unlike the mainstream route in Silicon Valley, Wu Yiming’s team starts from first principles, drawing on the embodied nature of human cognition, and finds an explainable mathematical framework.

The core of this technical path is to help machines understand distance, spatial relationships, and “common sense” that human toddlers can grasp through instinct—such as the fact that an object will bounce when it falls.

“Just like you can’t train a dog into a human being, so we tried to create genes for robots—writing into them gene code for recognizing and understanding the world the way humans do.” Wu Yiming used this analogy to describe the team’s exploration. “Based on this result, we give robots a brain and a soul, enabling them to understand space and the physical world, and link language symbols to physical objects—so in essence, we have created a new species, a silicon-based species.”

It is reported that the training cost for this technical route is almost zero, sharply contrasting with the current pattern of large models that consume enormous energy by default. Since the team was founded, it has cumulatively invested about 300 million yuan, and Wu Yiming estimates that the computing power and data investment required for the mainstream Silicon Valley route is, in effect, a “black hole.”

At present, this technology has been applied in major national missions, as well as in the fields of space and discrete manufacturing.

In the first half of this year, the team’s R&D “Xiaotian” robotic dog will be deployed for pilot trials in major scenic areas nationwide. These robotic dogs can autonomously recognize the environment, plan routes, pick up litter in scenic areas, and help carry items for visitors.

In the future, the team plans to launch three major product series: the intelligent visual technology VAI—robots’ “eyes”; an embodied intelligence operating system that is “plug-and-play”; and a brain module for embodied intelligence—integrating brain-like chips with integrated software and hardware—the embodied intelligence domain controller.

Wu Yiming expects that in the next 3 to 5 years, this technology will achieve rapid adoption across broad areas such as homes, healthcare, and transportation.

Regarding the information security concerns about artificial intelligence that the public widely worries about, Wu Yiming said that each embodied intelligence robot is equipped with an independent “brain,” with independent thinking and storage, which isolates from the source the risk of leakage caused by centralized cloud processing.

He also pointed out that current technology still has not reached the level where robots “meticulously scheme” to do something. The concern about “silicon-based species replacing humans” is too sci-fi, but risks that cannot be predicted must still not be ignored along the road to development.

A vast amount of information and precise interpretation—on the Sina Finance app