Unitree founder and chief executive Wang Xingxing says humanoid robot software is moving toward a major general-purpose breakthrough, while stressing that today’s systems remain limited. Speaking at the World Robot Conference in Beijing on August 20, Wang said the industry was approaching a “ChatGPT moment” for embodied intelligence. He used the comparison to describe a software threshold at which robots could enter unfamiliar environments and respond to ordinary voice or text instructions without extensive task-specific preparation.
A benchmark of 80% of tasks in an unfamiliar home
According to Reuters, Wang’s benchmark is a robot placed in an unfamiliar household that can successfully complete about 80% of requested tasks. Reaching that level could become a tipping point for rapid industry growth, he said. He did not present a firm timetable. In an optimistic scenario, a major advance in robot software could arrive within two or three years; a more cautious estimate put the range at five to ten years.
Wang also said Unitree’s largest current investment, measured by capital and personnel, is directed toward world models. These systems are intended to help physical AI represent surroundings and understand relationships among objects, actions and movement. In the same speech, he acknowledged that Unitree was lagging in real-world applications of physical AI models. That qualification matters because his forecast does not mean current robots already possess broad, general-purpose abilities.
The gap between demonstrations and useful work
The Associated Press reported that organisers expect about 3,000 products to be shown during the five-day World Robot Conference. At Unitree’s booth, robots danced, boxed and played table tennis with a human opponent. Other companies displayed industrial machines designed to move components in factory-like settings. Yet one helper robot observed by AP was unable to complete the apparently simple task of folding a shirt after several minutes of trying.
The demonstrations illustrate the gap between rapid progress in robot hardware and the dependable, adaptable software needed for everyday work. Reuters reported on August 19 that manufacturers at the conference were under pressure to move beyond attention-grabbing performances and demonstrate productivity, autonomy and customer value. Robots are entering selected factory and logistics settings, but humanoid systems remain less efficient than human workers in many applications.
China’s shipment volume has expanded
Reuters, citing a Chinese humanoid industry group, reported that China delivered more than 40,000 humanoid robots in the first half of 2026 and accounted for 97% of global shipments. The figures show rapidly expanding production scale, but they do not establish equally broad general-purpose adoption. Wang described the AI models responsible for decision-making and interaction as the industry’s main bottleneck. Companies at the conference are trying to move machines into factories, logistics operations and research settings.
Unitree’s market debut intensified attention
Wang’s speech followed Unitree’s heavily watched stock market debut in Shanghai. Reuters said the shares opened 11% lower on August 20 after rising nearly sixfold during their first trading session. The company’s listing prospectus indicates that universities and research institutions account for much of its customer base. That detail points to strong investor interest alongside a commercial market that is still developing.
Unitree is described as the world’s largest producer of robotic dogs by shipments and the second-largest producer of humanoid robots. Its public profile grew through choreographed movement and dance demonstrations, but Wang’s latest remarks shifted attention to software. His two-to-ten-year range is a forecast rather than a confirmed development schedule. The central message from the conference is that physical capability is improving quickly, while general-purpose software able to complete varied tasks reliably in unfamiliar settings is not yet ready.
