Unitree Robotics Develops Self-Evolving AI System for Humanoid Robots

According to founder Wang Xingxing, Unitree Robotics is working on a "self-evolving" feedback loop that would allow humanoid robots to autonomously write code, conduct simulations, and enhance their performance in the real world. He unveiled this initiative, named "Unitree Rehearsal," during a physical AI conference in Hangzhou on Thursday, highlighting it as the company's top priority for the year. This system aims to transition robotics development from laborious manual coding to a self-sustaining feedback loop. It would enable AI to explore research papers, generate code, and test it in simulations before applying it to physical robots. Although human oversight would remain for performance evaluation, Wang emphasized the goal of achieving an automated process to significantly speed up robotics algorithm development. The Hangzhou-based firm aims to minimize manual programming while leveraging AI advancements for practical robotics applications, reflecting a wider trend towards "self-improving" AI.