Unitree bets on 'self-evolving' AI to accelerate robot revolution

Unitree Robotics is developing a "self-evolving physical AI" system that could allow robots to accelerate their own development, its founder and chairman Wang Xingxing said on Thursday.

Wang's comments came a day after Unitree, one of China's best-known humanoid robot makers, made its debut on Shanghai's STAR Market, drawing fresh investor attention to the commercial prospects of embodied AI.

Speaking at the 2026 World Robot Conference in Beijing, Wang said the company was working on what it calls a "self-evolving physical AI robot" system designed to automate much of the robot development cycle.

Under the approach, advanced AI models could search research papers, write robot-control code, test it in simulated environments and then deploy it on physical machines. AI systems and human engineers would subsequently evaluate the results and feed the assessments back into the system, creating a closed-loop development process.

"Dramatically improving robot-development efficiency, and even enabling robots to evolve themselves, is something very worthwhile to pursue," Wang said.

If successful, the system could shorten development cycles by shifting parts of the labor-intensive process of programming, simulation, testing and optimization from human engineers to AI itself.

Wang said such "physical AI self-evolution" could eventually become a broader direction for the robotics industry, opening new opportunities for large-scale deployment as robots learn new capabilities more rapidly.

A major technological breakthrough could emerge within two to 10 years, he said, potentially bringing embodied intelligence to what he described as its "ChatGPT moment" — when robots become capable of handling a broad range of unfamiliar tasks rather than being trained for narrowly defined jobs.

Unitree has been pushing its robots into real-world environments, including automotive factories, Wang said. The company has also deployed robots in its own factories for simple testing tasks, while its AI team is experimenting with applications in household settings.

But Wang said Unitree had so far held back from large-scale deployment because robots still lacked the efficiency and general-purpose capabilities needed to compete with humans across a wide range of tasks.