Wang Xingxing of Unitree Robotics: Aligning models with real-world robots is the core challenge

Today, the 2026 World Physics-AI Conference and Unitree’s Partner Ecosystem Conference kicked off simultaneously in Hangzhou. Wang Xingxing, Founder, CEO, and CTO of Unitree Technology, stated in his keynote speech that the most critical barrier preventing embodied intelligence from reaching its “ChatGPT moment” lies in precisely aligning AI models’ input-output capabilities with the actual physical behaviors of robotic hardware. He elaborated further: for leading large-model companies, model self-evolution has already been largely achieved internally—models can autonomously write code and proactively collect data, reducing the human role to guidance and moderate participation. Meanwhile, Wang emphasized that it typically takes 5 to 10 years for a technology to diffuse from breakthrough to mainstream daily use—meaning external perception tends to lag. In the field of Physics-AI, however, clear signs of comprehensive breakthrough are already emerging.