Unitree builds Rehearsal AI to let humanoid robots improve themselves

Unitree founder Wang Xingxing unveiled Unitree Rehearsal, a system where AI searches papers, writes code and tests it in simulation before trials on real robots, then feeds results back into its models. He moved his timeline forward: humanoids should complete 80% of voice-set tasks by 2028.

• Rehearsal is Unitree's most important project of the year
• Humanoids should handle 80% of voice tasks by 2028
• Unitree currently makes tens of thousands of robots a year
• H2-X runs about 8 hours and jumps from 4 metres