Google DeepMind has launched two models of Artificial Intelligence (AI) this week that are aimed at robots smarter than ever. Instead of concentrating on keeping track of comments, the updated version of Gemini Robotics 1.5 and his Gemini Robotics-Eer 1.5 partner robots are thinking of problems, searching for information on the internet and sharing the skills between different robots agents.
According to Google, these models mark a “fundamental step that makes it possible to browse the complexity of the physical world with intelligence and agility.”
“Gemini Robotics 1.5 marks an important step for AI solution in the physical world,” said Google in the advertisement. “By introducing resources, we go beyond models that respond to assignments and make systems that can really reason, plan, use and generalize aids.”
And this term “generalization” is important because models have problems with it.
Robots equipped with these models can now perform tasks, such as separating clothing per color, making a suitcase based on weather forecasts found online or checking local recycling rules to throw waste correctly. Now, as a person, you can say, “So, so what?” But to do this, machines require a skill called generalization – the possibility of applying knowledge to new situations.
Robots – and algorithms in general – usually have difficulty with it. For example, if you learn a model to fold pants, it cannot fold a T -shirt unless engineers set each step in advance.
The new models change this. They can record signals, read the environment, make reasonable assumptions and perform tasks with different steps that were ever out of reach – or at least extremely difficult – for machines.
But better doesn’t mean perfect. In one of the experiments, for example, the team showed the robots a series of objects and asked them to place them in the right waste. The robots used their cameras to visually identify each item, consult the most recent recycling guidelines from St. Francis Online and then set up where they should be in themselves, just like a local person would do.
This process combines online search, visual perception and steps -by -step planning -making decisions based on the context that go beyond some older robots. The registered success rate was between 20% and 40% of the time; It is not ideal, but it is surprising for a model that these nuances could never have understood.
How Google Robots converts into super robots
The two models divide the work. Gemini Robotics-Eer 1.5 acts as the brain, discovers what needs to be done and creating a step-by-step plan. It has access to Google Search when you need information. Once the plan is defined, the natural language instructions passes on to Gemini Robotics 1.5, which ensures real physical movements.
In more technical terms, the new Gemini Robotics 1.5 is a Vision-Taal action model (VLA) that transforms visual information and instructions into motorcycle assignments, while the new Gemini Robotics-Eer 1.5 A Vision-Lining Model (VLM) creates several areas to complete a mission.
For example, when a robot separates clothing, it reasons internally on the task with the help of a line of thought: understanding that “separation by color” means that white clothing goes to a waste and color to another and then split the specific movements needed to take each item of clothing. The robot can explain its reasoning in simple language, making his decisions less than a black box.
Google CEO, Sundar Pichai, commented on the X and noted that the new models will enable robots to reason better, plan in advance, use digital tools as searching and learning from one type of robot to another. He called him “Google’s next big step to general user robots that are really useful.”
The launch functions Google, shared with developers such as Tesla, Figure AI and Boston Dynamics, although each company uses different approaches. Tesla focuses on mass production for his factories, with Elon Musk who promises thousands of units in 2026. Boston Dynamics continues to expand the boundaries of robot -harmful athletics with his background atlas. Google, in turn, races on an AI so that robots can adapt to any situation without specific programming.
The moment is important. American robotics companies stimulate a national robotics strategy, including creating a federal office aimed at promoting the sector at a time when China AI and intelligent robots make a national priority. China is the world’s largest market for robots that work in factories and other industrial environments, with around 1.8 million robots that are active in 2023, according to the Germany -based International Robotics Federation.
The DeepMind’s approach differs from traditional robot programming, whereby engineers carefully codes for every movement. Instead, these models learn from the demonstration and can adjust quickly. If an object slides out of the hands of a robot or someone moves something in the middle of a task, the robot adapts without hesitation.
The models are based on the previous work of DeepMind March, when robots could only perform unique tasks, such as opening a zipper or folding paper. Now they have to deal with sequences that many people would challenge – such as making up for a while after checking the weather forecast.
For developers who want to experiment, an approach is divided into availability. Gemini Robotics-Eer 1.5 was launched on Thursday via the Gemini API in Google AI Studio, which means that every developer can start building with the reasoning model. The action model, Gemini Robotics 1.5, remains exclusively for “selected” partners (ie “rich”, probably).
* Translated and edited with authorization of Decrypt.
In MB, your nomination is worth Bitcoin for you and your friends. For every friend who opens and invests an account, you get exclusive rewards. More information!
Google’s robots can now think, search the internet and only learn new tricks first appeared on Bitcoin Portal.
#Google #Robots #search #internet #learn #tricks #Bitrss #Crypto #World #News