Chinese robotics companies are trying to prove humanoid robots can do more than dance, fight and perform carefully rehearsed demonstrations, with this week's World Robot Conference putting real-world deployments firmly in the spotlight.
More than 300 mostly Chinese companies are attending the five-day event in Beijing, where over 2,000 exhibits and more than 150 product launches are expected. Humanoids sorting parcels, packing mobile phones and handling factory materials are among the systems on display as manufacturers push towards wider commercial use.
But even as robots move into warehouses and factories, one of China's biggest robotics executives says their intelligence still hasn't caught up with their physical capabilities. Unitree founder and CEO Wang Xingxing told the conference on Thursday that the industry is approaching a "ChatGPT moment" for embodied intelligence, while warning that the software breakthrough needed to get there could still be years away.
Humanoid Robots Move From Demonstrations To Deployment
China's humanoid robotics industry has become known for machines that can run, dance, box and perform martial arts. At this year's World Robot Conference, however, manufacturers are increasingly showing robots doing the less spectacular work required in factories and logistics operations.
Industrial humanoid robotics company Leju demonstrated robots moving and sorting crates and smaller objects. Zhang Dapeng, the company's assistant vice president, told Reuters that logistics is currently its most common use case, with Leju robots already being used to move boxes and load components at Chinese automotive plants and factories in Europe.
Robotera demonstrated a humanoid torso mounted on a wheeled base sorting parcels. The company said its technology has been deployed at China Post logistics sites since last year, with more than 100 parcel-sorting robots now operating across 15 warehouses in China.
DexForce, meanwhile, showed humanoid robots packing mobile phones into boxes. According to the company, the machines have been operating since earlier this year at a Lens Technology factory, where they can detect and correct problems such as phones being positioned at the wrong angle. Lens Technology supplies components to companies including Apple and Huawei.
Other manufacturers are pushing into different stages of deployment. Lumos Robotics is using robots to automate the assembly of key modules at its own factory and plans to expand them into final assembly. X Square Robot is testing robots in hundreds of homes this year as it prepares for gradual commercialisation in 2027. Its longest household deployment so far has lasted one month.
Despite that progress, large-scale adoption remains limited. Robots are beginning to replace human workers in some narrow manufacturing, logistics and service tasks, but many deployments are still pilots or operate within carefully controlled environments.
Unitree CEO Says Robot Intelligence Remains The Bottleneck
For Wang, the next major breakthrough won't come from making humanoid robots run faster or move more convincingly. It'll come from improving the AI systems that allow them to understand their surroundings and decide how to act.
"We are marching towards a 'ChatGPT moment' in embodied intelligence," Wang told the conference.
His benchmark for reaching that point is a robot that can enter an unfamiliar home and successfully complete around 80% of tasks using simple voice or text instructions. Unlike a machine trained to repeat a specific factory process, that would require a robot to understand and respond to situations it hasn't been explicitly prepared for.
Wang said reaching that level could become a tipping point for rapid growth across the robotics industry. However, he also cautioned that the necessary leap in robot software could still be two to three years away under an optimistic scenario, or take between five and ten years.
Galbot founder Wang He offered a similar benchmark at the conference, predicting that embodied intelligence could reach its "ChatGPT moment" by 2028. He defined the breakthrough as robots becoming capable of completing around 70% to 80% of everyday tasks without specialised training.
Unitree Pours Investment Into World Models
Much of the challenge comes down to the AI models controlling how robots understand and interact with the physical world.
Wang said Unitree's largest current investment in both capital and personnel is going into world models. These AI systems are designed to help robots build an understanding of physical environments so they can navigate them, respond to changes and choose appropriate actions.
He acknowledged that Unitree is still "lagging behind" in the real-world application of physical AI models and said limitations in the AI powering robot decision-making remain the biggest obstacle to mass deployment.
Those comments come during an extraordinary week for the company. Unitree made its Shanghai STAR Market debut on Wednesday, with its shares closing 460% above their IPO price and valuing the company at around $50 billion. The listing followed an IPO that was more than 8,000 times oversubscribed by retail investors.
The company had produced and delivered around 18,000 bipedal humanoid robots across its models by July, although many of its customers remain universities and research institutions rather than businesses using the machines in commercial operations.
The demonstrations taking place in Beijing show that commercial deployments are beginning to grow. But Wang's comments also draw a clear line between robots that can perform defined tasks in controlled environments and machines capable of adapting independently to the unpredictable world around them.
For now, closing that gap remains one of the biggest challenges standing between increasingly capable humanoid robots and mass adoption.
Comments ( 0 )