Google teach robots how to carry out master’s and master’s tasks that require more accurate – multimedia

Deepmind continues progress in robot training through artificial intelligence. Now, the company, which has been part of Google since 2014 Two new models for Amnesty InternationalStanding on No Gemini 2.0and Those who want to work as a basis for the next generation of automated assistants.

Technology begins to explain this The artificial intelligence models of robots need to meet three requirements to be useful, in fact, useful. The models should be able to adapt to different situations, to be interactive, and that adequate removal of interaction with organisms be canceled.

The first news is called Robotics Gemini, “represents a big step in performance in these three areas”The company says. The model uses the Gemini understanding skills to generalize specific situations and perform a variety of tasks, even if it is not trained on it.

Watch the video

Gemini robots is Ready to help robots deal with new objects and environments, as well as with different types of instructions. According to the company, the model is able to understand and respond to the natural language Adapting your behavior to the inputs received, but also for changes in the environment around you.

Click on the pictures to see in more detail

Complex tasksWith multiple steps or require an accurate treatment capacity of the objects, It is the last of the fields that Gemini robots prepare improvements.

The model has been trained from data from the Aloha 2 Robotic platform, and it can be displayed in other robots with arm. Google DeepMind states that the solution can be adapted to other types of more complicated majorityLike Apollo, developed by Apptronik.

Technology also reveals another model called Gemini robots. The advanced model of vision and language focuses on “Logical, embodied” capabilitiesand Improving spatial understanding of robots and helping them better interact with the world.

For example, by serving a cup of coffee to a robot equipped with the model, it will be able to explain the object and conclude what is the best way to interact with it without causing a “disaster”. According to the company, Gemini Robotics-A can make all the steps needed to control a robot without the need for additional settingsWith the success rate of two to three times from Gemini 2.0.