One step closer to complete Generative Artificial Intelligence

The AI ​​that reasons, plans and acts

The race for AIG artificial general intelligence has just won a new chapter and comes directly from the Google Deep Mind laboratories, as they reported, for the first time, an AI model that not only interprets commands, but also reasons, plans and acts almost autonomously in the physical world.

I'm talking about Gemini Robotics 1.5, a technology that can redefine the boundary between obedient machines and truly intelligent agents. The announcement presented not only one model, but two pillars that work together, the Gemini Robotics, capable of transforming vision, language and action into motor commands for robots and the Gemini Robotics ER, specialized in spatial reasoning and multi-stage planning.

Double the bet.

Together they function as brain and body with the promise of creating robots that think before acting, something that until now was restricted to the human domain. To understand the magnitude of this change, imagine the simple act of separating clothes by color; for a common machine, that would require rigid, limited and specific instructions. Instead, the Gemini not only understands the concept of colors, but also creates a logical sequence of steps.

Distinguishing shades, planning movements, deciding where to place each garment and even explaining in natural language why it is performing this action, it is as if the robot narrates its own reasoning before moving, the advancement is at the heart of the concept called VLA. Vision, language and action, which seeks to integrate sensory perception, language interpretation and physical execution in a single system.

“Knowledge” spreads between robots

Until now, models could convert commands and movements, but without real reflection or adaptation, the Gemini AI breaks that limit, reasoning internally before acting, dividing complex tasks into smaller, easier-to-execute segments. Most impressive, however, is not just in the execution, the Gemini showed the ability to learn between different robots. Skills trained on one platform like the Apollo humanoid or the Frenca robotic arm can be transferred directly to other models without needing to reprogram everything from scratch.

It is like teaching a human and seeing how knowledge spreads instantly to the rest of humanity, this flexibility opens the way to something greater, the possibility of generalist robots capable of working at home, in factories or in hospitals, learning continuously, without depending on pre-programmed instructions.

It is not just another technical advance, it is a sign that we are entering the era of intelligent physical agents, capable of thinking several steps ahead, evaluating risks, even aligning with security rules in real time.