Tech • AI • Robotics • Game
China’s New Bionic AI Robots Look Too Human With Artificial Skin
7/10TL;DR
China unveiled two life-sized humanoid news anchors in Beijing as companies in China, the United States and Europe pushed robots further into media, retail and adaptive sensing, while major limits on autonomy and useful deployment remained clear.
KEY POINTS
Humanoid anchors debut in Beijing
On November 5, at the World Media Summit in Beijing, Xinhua presented two full-scale humanoid news anchors, Xinyan and Xinyu. Built by UBTech, they were designed with artificial skin and humanlike facial motion intended to project calmness and trust. Their appearance was shaped by film director Zhang Yimou, who worked on facial expressions, eye movement and lip shapes.
Real-time eye contact, but still visibly robotic
The robots can face audiences directly, with a camera positioned where a pupil would be, allowing actual visual tracking. Reports from the event noted that despite the realistic faces, their movements remained mechanical, with regular head turns, blinking and eye darts that still set them apart from human presenters.
State-vetted AI powers the scripts
The anchors are linked to Xinhua Yuanxin, an AI system launched in June that draws answers from a library of current-affairs material already checked by Xinhua. The setup is meant to reduce AI hallucinations by limiting responses to approved material in real time. Xinhua has described the system as traceable, secure and controllable.
Accuracy and editorial control remain open questions
Restricting an AI to a vetted database may reduce fabricated claims, but it does not guarantee correct retrieval, summarization or translation. No public error rate has been released. The larger issue is editorial control: the same organization operating the anchors also decides what enters the knowledge library, with Xinhua previously emphasizing cultural and ideological security.
Not the first robot anchor, but a new combination
Robot news reading is not new. Japan demonstrated the childlike android Kodomoroid in 2014, and Xinhua itself launched a digital AI anchor in 2018. The difference now is the merging of a physical humanoid body with a question-answering news AI that can interact face to face rather than through a screen.
China tests robots in a live convenience store
In Shanghai on November 2, AgiBot put six G2 service robots through a 24-hour shift in an operating convenience store. The wheeled humanoid machines handled coffee, hot food, greeting, bagging, shelf checks and table cleaning during lunch, evening, overnight and breakfast periods, while more than 2 million people watched the livestream.
Strong headline numbers, mixed execution
AgiBot said the robots completed 1,730 tasks and served 683 customers. One robot made 147 coffees, another served 126 hot meals. But the trial also highlighted the gap between task execution and workplace judgment: one restocking robot reportedly struggled to act proactively, while a table-cleaning robot drifted off course and rolled out of the store after encountering an overflowing trash can.
Figure has many robots, but limited commercial work
In San Jose, Figure AI has amassed large numbers of Figure 03 humanoids across its headquarters, enough for observers to joke about whether the company conducts a nightly head count. Chief executive Brett Adcock said in June that robots already outnumbered people at the company, about 740 to 660. The factory had reached a pace of one robot per hour in April, and produced its 1,000th unit in July.
The industry can build robots faster than it can place them
Despite manufacturing progress, many humanoids still spend their time walking, charging, collecting training data or running endurance tests rather than performing paid work. Figure has cited deployments with BMW and Catalyst Brands, but has not disclosed robot counts. Across the sector, commercial demand still lags behind production capacity.
Estonian researchers create a robot that grows a new sensor
A team from the University of Tartu and the Italian Institute of Technology built a moth-inspired robot that can form a new light sensor after construction. Artificial blood carrying a chemical circulates through vein-like channels in its wings. When exposed to ultraviolet light, the material turns into a conductive plastic patch that acts as a functioning sensor, allowing the robot to react by flapping or lighting up.
Adaptive hardware shows promise, with major limits
The sensor became more sensitive with repeated UV exposure, creating a form of material memory that lasted more than four months in storage. But the system remains a proof of concept: the change is permanent, works through one chemical reaction, and still depends on human decisions about when and where the sensor should grow. The robot cannot yet decide for itself what new sense it needs.
CONCLUSION
Recent demonstrations show rapid progress in lifelike presentation, long-duration service work and adaptive robot hardware. They also underline the same unresolved challenge across the field: robots can increasingly perform actions, but still struggle with judgment, autonomy and trustworthy use in real-world roles.
Ask a question