- Wang Xingxing said robots could soon handle most everyday tasks in unfamiliar environments through natural language instructions
- The forecast is more aggressive than the five-to-ten-year timeline commonly cited across the robotics industry
Unitree (宇树科技) founder and CEO Wang Xingxing (王兴兴) said embodied AI could reach its “ChatGPT moment” within two to three years, when robots will be able to work directly in most unfamiliar environments and perform a wide range of everyday tasks.
Speaking at the opening ceremony of the 2026 World Internet Conference Digital Silk Road Development Forum in Xi’an on July 22, Wang defined the milestone as the point when robots can successfully complete around 80% of tasks in roughly 80% of unfamiliar environments using voice or text instructions.
“The remaining challenges are no longer fundamental scientific problems, but engineering ones,” Wang said, adding that the scale of engineering work required remains unprecedented.
This is not the first time Wang has forecast a near-term breakthrough in embodied AI.
Earlier this year, during the Nvidia GTC 2026 conference, he said the sector’s “ChatGPT moment” could arrive within one to two years, suggesting he continues to view general-purpose robotics as advancing faster than conventional industry expectations.
Wang noted that humanoid robots have progressed rapidly in recent years, evolving from basic walking to dancing, martial arts demonstrations and simple service tasks.
Unitree is now the world’s largest supplier of both quadruped and humanoid robots by shipment volume, while its H1 humanoid recently set a world record with a running speed of 10 meters per second.
Limitations on generalization
Despite the advances, Wang said today’s robots remain largely confined to structured environments, with generalization representing the industry’s biggest technical hurdle.
While robots can achieve near-perfect success rates on repetitive single-task operations, they still struggle to perform multiple tasks in unfamiliar settings.
He argued that large-scale, high-quality data collection will be key to overcoming that bottleneck.
Unitree has begun collecting full-body robot motion data at scale this year and plans to invest more than 2 billion yuan from its IPO proceeds into embodied AI foundation models.
Why it matters globally
Wang’s timeline is notably more optimistic than the five-to-ten-year horizon commonly projected by industry observers, reflecting growing confidence that embodied AI is approaching an inflection point.
If realized, it would mark the transition of AI from digital assistants to machines capable of operating in the physical world.
For global competitors, his comments also underscore China’s perceived advantage in rapidly generating real-world robot data through large-scale deployments and engineering iteration.
These two factors could accelerate progress toward more general-purpose robotic intelligence, possibly at a faster pace than the rest of the world, as some observers predict.
