Unitree’s Wang Xingxing: Robot Core Bottleneck Is Millimetre-Level Error

Wang's framing shifts the bottleneck from model capability to physical-world calibration — a millimetre gap that no amount of scaling alone closes. That matters because it recasts the sector's timeline around hardware and control precision, not just AI research. His 80/80 benchmark is a useful, falsifiable yardstick for judging progress, though it is a prediction rather than a measured threshold. Whether the industry converges on shared evaluation standards for unfamiliar scenarios is the open question.