Wang Xingxing of Unitree: Signs that physical AI is about to be broken through have appeared, and it could happen in as soon as 1 to 2 years
Wang Xingxing, founder of Unitree Robotics, said at the 2026 World Physical AI Conference that signs that physical AI is about to be broken through have already appeared, and it could happen in as soon as 1 to 2 years. He pointed out that robot planning and navigation have been solved fairly well; the only issue not yet fully solved is the alignment between model inputs and outputs and the real physical world—for example, the success rate of grasping unfamiliar objects declines. Unitree is working to complete a self-evolution loop for physical AI robots: the model writes code itself, collects data, and trains in a test environment, with humans only needing to give occasional feedback.
Body · AI translation
Article body EM_StockImg_Start
EM_StockImg_End
On October 22, the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference, hosted by Unitree Robotics, was held in Hangzhou. Wang Xingxing, founder, chairman and CEO of Unitree Robotics, once again expressed regret in his keynote speech that he had not believed in AI earlier. He also pointed out that physical AI may be achieved by frontier AI as early as next year.
"To be honest, I think everyone in the world—including myself—still does not attach enough importance to and have enough faith in AI. I was just thinking that currently, perhaps only a few thousand people worldwide truly understand the potential of cutting-edge AI models. Most people simply do not know where the real boundaries of AI's capabilities lie," Wang Xingxing said.
He also said that signs of a breakthrough in physical AI have already appeared, and it is very likely to happen in the next 1 to 2 years, or 2 to 3 years.
"At present, the planning and navigation problems of robots have been solved fairly well. The only thing not yet fully solved is the alignment between model input and output and the real physical world. For example, when a robot is asked to grasp an unfamiliar object it has never seen before, the success rate drops. This is the core hard problem of physical AI today. But we are working through a path: self-evolution of physical AI robots. The model writes code itself, collects data itself, and trains in a test environment. Humans only need to occasionally give it a push, check a box, or provide feedback, and the model then iterates again based on that feedback. This closed loop is gradually being established," he said.
Wang Xingxing also pointed out that Unitree Robotics is committed to working with ecosystem partners to advance global robotics technology.
"Since the second half of last year, everyone can see that many robots, both in China and overseas, can perform very good movements and shows. Why? Because everyone has developed many excellent algorithms and open-source frameworks on Unitree's platform, and then developers around the world continue to iterate on that basis," he said.
(Source: 21st Century Business Herald) Body middle embedded advertisement