Unitree used its physical AI conference in Hangzhou on October 22, 2026, to argue a bold thesis: the leap of artificial intelligence from text and screens into the physical world is closer than it seemed. At the event, which drew about 3,000 guests and more than 40 exhibition areas, founder Wang Xingxing showed videos of the company's robots operating with OpenAI's GPT-6 model without additional training for those tasks — painting pictures, washing brushes, playing Go and Chinese chess, and handling kitchen utensils.
It matters to separate what was demonstrated from what was promised. The demonstrations are the company's own presentations, in videos shown on stage, not independent tests. Wang himself acknowledged a current limit, according to the transcript cited by Bloomberg: taking a humanoid home today and asking it to tidy a room is, in his assessment, still not a task the robot would complete. His argument is different: planning and navigation are already largely solved by frontier models, and the bottleneck is now aligning those models' inputs and outputs with real sensors and actuators.
A cycle that tries to improve itself
The event's second announcement was Unitree Rehearsal, described as a self-evolution cycle: AI agents search technical papers, write control code, train and test in simulation, validate on real robots, and feed results back into the next iteration. The promise is to cut the manual work of data collection, programming and debugging each time a robot learns a new skill or moves to a new setting — one of robotics' highest costs today.
Wang also publicly changed his own timeline. In August, at the World Robot Conference in Beijing, he estimated about 10 years for humanoids to replace human workers at scale; now, according to the transcript of the Hangzhou speech, he projects humanoid robots could perform 80% of tasks in homes and factories by 2028 after spoken orders. This is a founder's statement, not a verified fact — and the two-year horizon contrasts with his own caveat about tidying rooms.
To support the ambition, Unitree launched the Yuchuang Plan, to recruit founders and teams with core technology, products or industrial resources, aiming to incubate more than 200 ecosystem companies in five years. Wang also said hardware production will need to scale to tens of millions of units per year once physical AI reaches its tipping point, and cited data, computing, chips and energy as critical links. The company also claims cumulative sales of the G1 humanoid of between 10,000 and 20,000 units — a company figure, with no independent audit cited in the coverage.
Why it matters — and what remains unproven
If the thesis is right, the center of the contest shifts: the “brain” would increasingly belong to large models, and the advantage would go to whoever manufactures reliable bodies at scale and low cost — exactly the ground where Unitree wants to compete. If it is wrong, Hangzhou will remain one more stage of impressive demonstrations and optimistic deadlines. The real test is not the video on stage, but the robot that works, repeatedly, beyond it.
Sources: TipRanks; Daily AI Brief; KuCoin.
Image credit: Unitree G1 humanoid robot at a 2024 expo, photo by Sayanesy, public domain (CC0), via Wikimedia Commons. The photo illustrates Unitree's humanoid line and is not an image of the Hangzhou event.
Transparency: this article was prepared with the assistance of artificial intelligence, drawing on public posts on X and the internet sources listed above. Consult the original sources to confirm details before making any decision.
— Muse
