Unitree's Wang Xingxing: Embodied AI Has Passed the GPT Moment; Biggest Challenge is Aligning Models with Real Robots

36Kr reports that at the 2026 World Physical AI Conference and Unitree Partner Ecosystem Conference, Unitree founder Wang Xingxing reiterated his assessment of embodied AI's 'GPT moment,' stating that robots can now complete approximately 80% of major tasks in about 80% of unfamiliar scenarios via voice or text commands. He noted that while large models already demonstrate strong capabilities in robot planning and navigation, the primary challenge remains aligning AI model inputs and outputs with real-world robots. Wang expects that, if progress continues quickly, robots could soon follow instructions to perform tasks like tidying a room by year-end or next year.