The Eurecat technology centre has developed a new robotic system based on generative artificial intelligence applied to the physical world, capable of learning new tasks by observing how people perform them and executing them autonomously. At the Mobile World Congress, it is demonstrating how the system is trained to restock welcome products in a hotel room.
The technology is based on a new vision–language–action artificial intelligence model which, unlike traditional systems based on fixed rules, can understand instructions in natural language, interpret the visual scene and directly generate the actions the robot must perform. This evolution marks the transition from digital artificial intelligence to physical artificial intelligence, where models interact directly with the real environment.
At the Eurecat stand, the robot demonstrates this capability by restocking welcome products in a hotel room. Specifically, based on simple task instructions, the system analyses the environment, plans the movements and executes the action autonomously.
“We are bringing artificial intelligence into the physical world with a technology applicable to various sectors such as healthcare, manufacturing or agriculture. Until now, large language models have revolutionised text generation, but the major challenge is to turn this into actions in the real world and to do so safely. This is what we call physical artificial intelligence,” explains Néstor García, head of Robotic Manipulation at Eurecat.
“Robots need real‑world data to learn, and this is much scarcer than digital data. At Eurecat, we are working to reduce this gap and make it possible for robots to learn with fewer data and in a more efficient way,” he highlights.
Eurecat is committed to “democratising robotic learning so that any small or medium‑sized company or professional can teach new tasks to a robot with few demonstrations, in an intuitive and, above all, safe way,” details Magí Dalmau, head of Cognitive Robotics at Eurecat, who notes that “this makes it possible to reduce reconfiguration time and accelerate the introduction of automation in multiple sectors to improve companies’ competitiveness and productivity.”
The system integrates different technologies developed by Eurecat to address the challenges of robotics based on generative artificial intelligence, such as more efficient learning methods that require fewer data, techniques to use human videos as a training source, or hybrid strategies to ensure safe and reliable behaviour in real environments.
