Skip to main content

Wang Xingxing: Embodied Intelligence Could Hit Its 'ChatGPT Moment' in 2-3 Years

Wang Xingxing: Embodied Intelligence Could Hit Its 'ChatGPT Moment' in 2-3 Years

At the 2026 World Robot Conference, Wang Xingxing, founder and CEO of Unitree Robotics, took the stage with a bold prediction: embodied intelligence is on the verge of an industrial tipping point, much like the moment ChatGPT reshaped the AI landscape. He estimated that this breakthrough could come in as little as 2 to 3 years, or up to 5 to 10 years in a more conservative scenario.

The timing was notable—just a day earlier, Unitree had listed on Shanghai's Science and Technology Innovation Board, with its stock price soaring 460% on the first trading day. Wang's confidence isn't just about market enthusiasm; he laid out a clear benchmark for when embodied intelligence truly enters the industrial boom: robots must be able to complete about 80% of tasks through voice or text commands in 80% of unfamiliar environments. That's a far cry from today's reality, where robots often struggle when conditions shift even slightly.

The Last Few Centimeters Problem

Wang didn't shy away from the industry's biggest headache: aligning AI models with the messy, unpredictable physical world. In controlled, fully trained settings, robots can achieve high success rates. But change an object's position, alter the lighting, or introduce a new obstacle, and those success rates plummet. The 'last few centimeters'—or even millimeters—of manipulation remain a stubborn hurdle, with tactile feedback and action errors limiting generalization.

This is where Unitree's 'Physical AI Robot Self-Evolution V1.0' roadmap comes in. The idea is to use AI large models to automatically retrieve research findings, generate control code, and then iterate through simulation training, real-machine testing, and feedback from both AI and human evaluators. Wang believes that the foundation model's capabilities, multi-source data accumulation, the scale of real robot deployment, and skill building will be the key drivers of this self-evolution.

From Quadrupeds to Humanoids

Wang also took a moment to reflect on Unitree's journey, from quadruped robots to the humanoid G1, which has become an industry icon—even performing at the Spring Festival Gala. This year, the company expanded its lineup with a manned exoskeleton and the wheeled-legged robot As2-W, pushing into factory, home, and outdoor applications.

But Wang was candid about the road ahead. To achieve large-scale deployment, robots need to solve efficiency and generalization issues. Right now, they can handle simple tasks, but new tasks often require retraining from scratch. As AI models improve and real-world data accumulates, embodied intelligence should accelerate into industrialization.

A Shift in Mindset

The industry is moving from 'manufacturing robots' to 'enabling robots to continuously learn and evolve.' Data, models, hardware, and self-evolution capabilities are becoming the new battlegrounds. Wang's vision suggests that the next few years will be critical—not just for Unitree, but for the entire field of embodied intelligence.

Key Points

  • Wang Xingxing predicts embodied intelligence will reach a 'ChatGPT moment' in 2-3 years.
  • The benchmark: robots must handle 80% of tasks in 80% of unfamiliar environments.
  • The main bottleneck is aligning AI with the physical world, especially in fine manipulation.
  • Unitree's 'Physical AI Robot Self-Evolution' roadmap aims to create a continuous learning loop.
  • The industry is shifting from building robots to enabling them to evolve.