In August, Anhui Baize Tongjing Technology Co., Ltd. (hereinafter referred to as Baize Tongjing) is set to secure its first round of financing.

It is a technology start-up founded by a second-year graduate student in artificial intelligence at the University of Science and Technology of China (USTC), with all members of its technical team born in the 2000s. Not long ago, at an enterprise roadshow held as part of the Anhui leg of the 2026 “Vibrant China Research Tour,” Bai Yinqi, the 24-year-old founder, CEO and CTO of Baize Tongjing, introduced the team’s cutting-edge research focus—world models—to media reporters from across the country.

On August 11, at Baize Tongjing, located in Grand Union of Innovation, a robot equipped with a small black box was performing motion calculations. The seemingly unremarkable box is the “brain” for embodied intelligence developed by the team.

“Many of the robots we see dancing or performing rely largely on pre-programmed instructions, which are essentially the robot’s ‘cerebellum,’” Bai Yinqi explained. By contrast, a world model functions as the robot’s “brain.” By processing visual input from the surrounding environment and learning physical laws, it enables robots to determine what to do next autonomously. “For example, rather than having a robot memorize the trajectory of a falling apple, we want it to understand the law of gravity. That way, whether it is an orange or a stone, the robot can still make the right judgment,” Bai said, giving an example.

To this end, the team has developed a technical approach combining latent space, causal reasoning and scaling with unlabeled videos, with practical application as its top priority. During the pre-training stage, massive amounts of online video are used to enable the model to learn physical laws autonomously, without extensive manual annotation of actions. More costly and scarce real-world robot data are used only in limited amounts during the later alignment stage, significantly reducing labor and data costs. In the future, this “robot brain” can be applied not only to humanoid robots but also extended to drones, autonomous-driving systems and other intelligent systems.

Source: Hefei Daily