Unitree developing AI system to let humanoid robots improve themselves

Unitree Robotics' "Unitree Rehearsal" project is described as a "self-evolving" feedback loop intended to enable humanoid robots to enhance their performance autonomously. This initiative, which founder Wang Xingxing called the company's most important of the year, aims to shift robotics development away from intensive manual programming.

The proposed system involves an AI that would search research papers, write code, and then test these developments in simulations before conducting trials on physical robots. This structured, iterative process is designed to create a continuous, autonomous cycle for improvement, reducing the need for direct human intervention in each step of algorithm development.

Unitree's ultimate goal is to establish an automated cycle capable of exponentially accelerating the development of robotics algorithms. While the system is designed for independent operation, Wang noted that humans could still provide assistance in assessing performance and offering feedback, suggesting a potential role for human oversight within this advanced automation framework.