Innovative Study on Robot Training by Google DeepMind

The team at Google DeepMind responsible for robots is developing a new approach in the training of AI-powered robots. The team has published an impressive article showing that the RT-2 robots, operating with the Gemini 1.5 Pro AI model, have learned by watching videos.

The Gemini 1.5 Pro model provides a window for training robots through long-form videos, enabling robots to process a large amount of information at once. Researchers capture a video tour of a specific area to give robots the ability to recognize their surroundings and perform tasks. The details in the video tour help robots successfully complete their tasks.

This study can be considered an important step in enhancing robots’ abilities to interact with their environments and perform complex tasks. Additionally, the publications by Google DeepMind demonstrate that the success of this technology is not accidental. Gemini-supported robots can perform various tasks in real life with a high level of accuracy.

One of the most remarkable features of the Gemini 1.5 Pro model is its ability to successfully complete multi-step tasks. Robots have reached a level where they can, for example, go to a refrigerator, review its contents, and determine whether a specific drink is available or not.

Although this technology is not yet at a consumer-ready level, AI-powered robots are expected to play a significant role in the future in the healthcare, transportation, and cleaning sectors. Models like Gemini 1.5 Pro represent a significant advancement in robot technology.