Google DeepMind has introduced two AI models, Gemini Robotics and Gemini Robotics-ER, designed to improve robot control.

Gemini Robotics integrates vision-language-action (VLA) capabilities, enabling robots to interpret visuals, understand language commands, and perform movements with enhanced dexterity and interactivity.

It excels at tasks requiring fine motor skills and adapts to various environments.

Gemini Robotics-ER focuses on embodied reasoning, improving spatial awareness and allowing robots to integrate with existing control systems.

It enables robots to perform more complex tasks and interact with objects more naturally.

Both models emphasize safety, with features like collision avoidance and the ability to assess actions’ safety, Interesting Engineering has reported.

DeepMind is also releasing a dataset for evaluating AI safety and collaborating with industry partners to create safer, adaptable AI-driven robots.

Written by B.C. Begley