rewrite this title Google DeepMind’s New AI Model for Robots: Gemini Robotics

Here’s a grammatically correct English translation of the provided text, with key terms bolded:

Google DeepMind’s New AI Model for Robots: Gemini Robotics

Gemini Robotics and Gemini Robotics-ER enable robots to interact with objects and navigate environments in the real world. Google DeepMind announced its new artificial intelligence models, Gemini Robotics. These models allow real-world machines to interact with objects and navigate environments. DeepMind shared videos of robots equipped with Gemini Robotics performing various tasks, including folding paper and placing items into a bag in response to voice commands.

According to DeepMind’s sharing, Gemini Robotics was trained to generalize behavior across a range of different robotic hardware. The training focused on associating items that robots “can see” with actions they can perform. DeepMind stated that Gemini Robotics enabled robots to perform well in tests even in environments not included in their training data.

Furthermore, the institution also released a scaled-down model called Gemini Robotics-ER. Researchers can use this model to train their own robotic control models. Both Gemini Robotics and Gemini Robotics-ER run on Gemini 2.0. DeepMind also released a metric called Asimov to measure risks with AI-powered robots.

Apptronik Partnership

You may recall that we recently mentioned Apptronik when discussing the top 5 humanoid robots that have recently gained prominence. Google stated that it will partner with Apptronik to build next-generation humanoid robots with Google Gemini 2.0. It’s worth remembering that Google was also part of Apptronik’s $350 million investment round. In the promotional videos shared by Google, it’s possible to see Apptronik robots equipped with the new AI models. The robots are shown filling a lunchbox or moving plastic vegetables in response to verbal commands. The release date for this technology to the market remains uncertain for now.

You Might Also Like;

Copy URL
URL Copied