Google robots get smarter after integration with Gemini AI

Robots used in Google offices systems became smarter after being optimized with Gemini featuresThe company’s productive artificial intelligence. This was revealed in a study published by experts from the technology giant on Wednesday (10).

According to the DeepMind robotics team, the machines were trained with the wide context window of Gemini 1.5 Pro, the most powerful version of the language model. The idea was that: Developing advanced robotic navigation by combining the processing powers of AI with computer vision.

And The results were considered promisingfacilitates interaction between humans and machines. In one experiment seen in the video above, someone asks a Gemini robot to help him find a place to draw, and begins the conversation with the phrase, “Hey, robot.”

The robot then reports that it has processed the request with the help of generative AI and after a short while directs the user to the whiteboard on the wall. According to Google, can do even morehelps a person move more efficiently in the environment.

Points for improvement

Those conducting the research stated that they were successful Success rates of 86% to 90% on previously impossible navigation tasksFeaturing complex reasoning and multimodal instructions across more than 50 interactions. The experiment was conducted in an area of ​​835 m², an area that the machine already knew.

However, there are still some areas for improvement, such as the time it takes for the robot to process the request with Gemini and formulate the response. Varies from 10 to 30 secondsAnother detail is that the machine cannot yet perform its own demonstration drive autonomously.

The researchers also want to find out whether the Gemini 1.5 Pro can steer robots. plan and execute more complex tasksin future studies. Preliminary evidence for this capacity has already been identified.