Unitree's Wang Xingxing: Millimeter Errors Are Robotics' Core Bottleneck

On September 24, the fifth Global Digital Trade Expo opened in Hangzhou, where Unitree Technology founder Wang Xingxing delivered a keynote titled "From Machinery to Intelligence — An Evolution Theory of the Embodied Future," sharing his views on how embodied intelligence will develop.

Reviewing the AI industry's trajectory, Wang noted that ChatGPT fundamentally reshaped public understanding of artificial intelligence and expanded the sector's imagination. On that basis, he predicts the embodied intelligence field will likewise get its own "ChatGPT moment."

He set a clear threshold for that moment: when robots can complete 80% of tasks in 80% of unfamiliar scenarios through voice-driven embodied capabilities, the industry will have reached its tipping point. At that stage, companies and national resources worldwide would pour into the sector, a turning point he expects to arrive within the next few years.

Wang said that getting robots to follow instructions and symbolically perform a specific job was already achievable last year. The biggest problem today, he argues, is that AI models' inputs and outputs do not match the real physical world precisely enough, so robot work carries errors of a few millimeters. Whoever solves that problem, he says, solves robotics entirely.

Wang made a similar point during the 2026 World Robot Conference in August, where he explained why Unitree's products have not been rolled out at scale in factories or homes. The main reason is that overall efficiency and capability remain insufficient — robots can do some work, but they are still slower than humans.

He added that humanoid robots currently must be retrained whenever they face a new task, which drags efficiency down further. The team therefore hopes to make the technology more general and its capabilities more precise before pursuing large-scale deployment in specific scenarios.