At the "Internet Entrepreneurs Forum" of the 2026 World Internet Conference Wuzhen Summit, Wang Xingxing, founder of Unitree Robotics, said that embodied AI is expected to reach a breakthrough moment similar to ChatGPT, when robots will be able to complete 80% of tasks through voice commands in 80% of unfamiliar scenarios. At present, top models are already very close to this goal.
Wang Xingxing pointed out that current embodied AI models already have strong planning capabilities, but in fine operations such as disassembly and assembly, they still face challenges in physical alignment and tactile contact—the "last few millimeters" problem. He believes that 2027 may become a tipping point for the generalized deployment of embodied AI. Once this technical bottleneck is overcome, the barriers to large-scale industry application will be greatly reduced.
In addition, several months ago Unitree Robotics began internally exploring the self-evolution of physical AI models, allowing models to write code autonomously, train in simulated or real environments, then deploy them to robots for testing, and continuously revise and optimize programs based on feedback from humans or robots. Wang Xingxing said that the industry's robot data utilization rate is still low. As foundation model capabilities improve, data utilization efficiency and model self-evolution capabilities are expected to be further enhanced.
As for future development directions, Wang Xingxing said that Unitree Robotics will focus on conquering robot brain technology and seek cooperation globally to drive breakthroughs in physical AI technology.
