- The new AI model of Google Gemini Robotic
- The model can learn new tasks from only 50 to 100 examples
- It adapts to different types of robots, such as humanoids or industrial arms, and could be used in rural houses and hospitals
For years, we have been promised robot butler capable of folding your laundry, cutting your onions and trying a good engine full of mind like those of our favorite dramas. One thing that these promises never mention is that the accidentally disconnection of your router could close these mechanical jeeves. Google claims, however, that its new Gemini AI model solves this problem.
Google Deepmind has unveiled its new gemini robotics intermediary model on AI devices as a means of protecting robots from shot down and working in rural areas. Although it is not as powerful as the standard -based geminum models based on the cloud, its independence means that it could be much more reliable and useful.
The breakthrough is that AI, a VLA model (vision, language, action), can look around, understand what they see, interpret the instructions of natural language, then act on them without having to search for words or tasks online. During the tests, the robots with the installed model were completed on unknown objects and in new environments without googling it.
It may not seem to be a huge affair, but the world is full of places with the internet limited or no access at all. Robots working in rural hospitals, disaster areas and underground tunnels cannot afford to drag. Now, not only is the model quickly, but Google claims that it has an incredible ability to learn and adapt. The developers say they can teach new AI tips with as little as 50 demonstrations, which is practically instant compared to some of the programs currently used for robotic training.
AI offline robot
This ability to learn and adapt is also obvious in the flexible physical design of the robot. The model was first designed to execute Google’s own double -arm Aloha devices, but has since proven itself capable of working when it is installed in much more complex machines, such as the Apptronik Apollo Humanoid robot.
The idea of machines that learn quickly and act independently, obviously raises certain red flags. But Google insists that it is careful. The model comes with integrated guarantees, both in its physical design and in the tasks it will perform.
You cannot miss and buy a robot with this installed model, but a future involving a robot with this model or one of its descendants is easy to imagine. Let’s say you buy a robot assistant in five years. You want it to do normal things: fold the towels, prepare meals, prevent your toddler from launching Lego bricks on the stairs. But your other child wanted to see how this box with the flashing lights worked, and suddenly, these lights stopped flashing. Fortunately, the model installed in your robot can always see and understand what these Lego bricks are and that you ask him to pick them up and put them back in their bucket.
This is the real promise of gemini robotics on devices. It is not only a question of bringing AI to the physical world. It is a question of making it stay when the lights sparkle. Your future butler robot will not be a responsibility for a cloud. The robots arrive and they are really wireless. I hope it’s always a good thing.
