A humanoid robot standing in a research lab next to workstation monitors showing abstract robot simulations (AI-generated image)
AI-generated image

Unitree

Unitree says it is building "Rehearsal", a self-improving loop for robot software

By Humanoid News Staff · Published · Last updated

Short answer: Unitree founder Wang Xingxing said on 22 October 2026 that the company is developing "Unitree Rehearsal", a "self-evolving" feedback loop designed to let humanoid robots write code, run simulations and improve their real-world performance with less manual programming, the South China Morning Post reported.

SCMP says Wang unveiled the project at a physical AI conference hosted by Unitree in Hangzhou and called it the company's most important initiative of the year. In the system under development, AI would search research papers, write code and test it in simulation before running trials on physical robots.

A humanoid robot walking along a marked practice course in a robotics lab
Scene: a humanoid on a practice course in a lab. AI-generated image

Humans could still assess performance and give feedback, according to SCMP, but Wang said the ultimate goal is an automated cycle that could greatly speed up the development of robotics algorithms.

A desk with computer screens showing wireframe humanoid skeleton simulations
Scene: wireframe simulations of a humanoid on workstation screens. AI-generated image

PANews, summarising his keynote, reports that Wang said leading AI-model companies have in a sense already achieved self-evolution of models internally, and that the core challenge for embodied AI to have its "ChatGPT moment" is aligning AI model inputs and outputs with real robots. Neither report gives technical details, test results or a release date.

← All news