A few months ago, Unitree began exploring in-house AI models that can evolve on their own to control robots, according to Wang Xingxing. He showed a preliminary version of this research project on stage, called “self-evolution of physical AIAI systems that perceive, understand and act in the physical world: robots, autonomous vehicles, smart cameras.Read the glossary entry robots V1.0”, CLS reports.
The chief executive described the principle. The model writes its own code and trains in simulation or in real conditions. The result is then installed on a robot to be tested. The success rate of that test is used to correct the code before further training.
The role of humans would change under this scheme. “For now, I think there is only one major use left for humans: giving scores,” said Wang Xingxing. In his view, humans would evaluate the models and set goals for them.
The founder and CEO of Unitree was speaking on 4 November at a round table for internet entrepreneurs, at the Wuzhen summit of the World Internet Conference 2026.
Source
- 钛媒体 TMTPost · tmtpost.com
Also worth reading
Artificial intelligence
Rhoda AI wants robots to learn from web video, not just teleoperation
Illustration, not a photo of the robot