Google DeepMind Unveils New AI Models for Robot Control

Gemini Robotics: Google DeepMind's New AI Models for Robotics

Google DeepMind, the artificial intelligence research division of Google, unveiled a new suite of AI models on Wednesday known as Gemini Robotics.

These models are specifically engineered to empower physical machines with the ability to interact with objects and effectively navigate their surroundings.

Demonstrating Real-World Capabilities

A series of demonstration videos released by DeepMind showcase robots utilizing Gemini Robotics to accomplish various tasks.

These tasks, executed in response to spoken instructions, include folding paper and carefully placing eyeglasses into their protective case.

The lab reports that Gemini Robotics has been trained to adapt its behavior across a diverse spectrum of robotic hardware configurations.

Furthermore, the system establishes a connection between visual input – what the robot “sees” – and the appropriate actions to be taken.

Generalization and Accessibility

DeepMind asserts that testing has demonstrated the ability of Gemini Robotics to perform effectively in environments that were not part of its initial training dataset.

To facilitate further research and development, a streamlined version of the model, Gemini Robotics-ER, has been made available.

This allows researchers to train their own customized robotics control models.

Introducing the Asimov Benchmark

Alongside the models, DeepMind has also introduced Asimov, a new benchmark designed for evaluating and mitigating potential risks associated with AI-driven robotics.

This benchmark will aid in the responsible development and deployment of increasingly sophisticated robotic systems.

Asimov provides a framework for assessing the safety and reliability of AI-powered robots.