Unitree Founder Identifies Millimeter Precision Gap as Core Embodied AI Bottleneck
Speaking at the Global Digital Trade Expo, Unitree Robotics founder Wang Xingxing stated that the primary technical bottleneck for embodied AI is bridging the gap between AI model outputs and the physical world to eliminate millimeter-level execution errors. He predicted that embodied AI will reach its inflection point—its 'ChatGPT moment'—when robots can independently complete 80% of tasks across 80% of unfamiliar environments. Achieving millimeter-level precision and high task generalization is essential before humanoid robots can transition from controlled demos to widespread deployment in factories and homes. Overcoming these physical execution errors will trigger massive global investment and mark the commercial tipping point for the robotics industry. Wang noted that current robots suffer from lower efficiency compared to human workers and require time-consuming retraining whenever presented with new tasks. Rather than rushing into immediate large-scale industrial or household adoption, Unitree is prioritizing foundational AI precision and generalization capabilities.
## BACKGROUND
Embodied AI refers to artificial intelligence agents operating within a physical body—such as humanoid or mobile robots—that interact directly with the physical environment. Unlike text- or software-based AI models like ChatGPT, embodied AI must continuously process sensor inputs, make decisions, and execute physical movements in real time under real-world constraints.