Introduction to Gemini Robotics

On Wednesday, Google DeepMind, the AI research lab of Google, introduced a new range of AI models known as Gemini Robotics, which is designed to allow real-world machines to interact with objects, navigate through different environments, and perform various tasks.

Demonstration and Capabilities

DeepMind has published a series of demonstration videos showcasing robots equipped with Gemini Robotics performing tasks such as folding paper and placing a pair of glasses into a case in response to voice commands. According to the lab, Gemini Robotics has been trained to generalize its behavior across a wide range of robotics hardware and to associate items that robots can perceive with potential actions they might take.

Performance and Availability

DeepMind reports that Gemini Robotics has shown impressive performance in environments that were not included in its training data. Additionally, the lab has made available a streamlined model called Gemini Robotics-ER, which allows researchers to train their own models for robotics control. Furthermore, DeepMind has introduced a benchmark known as Asimov, designed to assess the risks associated with AI-powered robots.