Unitree develops AI system to help humanoid robots improve themselves
Founder Wang Xingxing expects humanoids to handle 80% of voice-assigned tasks as soon as 2028, bringing forward his outlook for practical robots.
Unitree Robotics is developing an AI system intended to help its humanoid robots improve their own capabilities, founder Wang Xingxing said at a company-hosted conference in Hangzhou. Wang described the project, called Unitree Rehearsal, as the company’s top priority for the year, the South China Morning Post reported.
The proposed system would use AI to review research, produce code and evaluate it through simulations before testing on physical robots, according to the South China Morning Post. Bloomberg reported that Unitree already uses AI for this process, feeding results from physical tests back into its models.
Wang now believes humanoids could perform 80% of tasks assigned through voice commands as soon as 2028, Bloomberg reported. The forecast includes factory and household work, according to a Unitree-provided transcript cited by Bloomberg. At the World Robot Conference in Beijing in August, he had suggested widespread adoption in those settings could take 10 years.
Wang attributed his revised outlook to advances in AI models. He said Unitree is developing models of its own while collaborating with Chinese and overseas AI labs. Bloomberg also reported that the company plans to increase manufacturing capacity for potential demand reaching tens of millions of robots, compared with current annual output in the tens of thousands.
