Google

Total Body Mastery: How Google DeepMind’s AI Controls a Robot

Published

56 minutes ago

July 30, 2026

Google DeepMind’s new AI model can control a robot’s entire body

Google DeepMind Introduces Gemini Robotics AI Model for Humanoid Robots

Google DeepMind has announced the latest version of its Gemini Robotics AI model, which is capable of controlling entire humanoid robots. The previous model focused on controlling the upper body of humanoid robots, but Gemini Robotics 2 now enables “whole-body motions” from feet to fingertips.

Expanded Range of Actions for Humanoid Robots

The new Gemini Robotics model allows humanoid robots to perform a wider range of actions, including walking, crouching, stretching, and manipulating objects. Videos shared by Google demonstrate how Apptronik’s Apollo 2 robot can pick up a watering can, as well as retrieve specific items from a shelf.

Enhanced Dexterity and Coordination

Gemini Robotics 2 now supports better dexterity with the ability to control complex, five-fingered hands. This advancement enables robots to perform tasks such as sealing a Ziploc bag, tying a trash bag, or unscrewing a lightbulb.

Updates to Gemini Robotics ER for Embodied Reasoning

Google DeepMind is also updating Gemini Robotics ER (embodied reasoning), a vision-language model that assists robots in analyzing their surroundings, following instructions, and executing multi-step tasks. Gemini Robotics ER 2 is now more adept at completing tasks over extended periods and has a better understanding of task initiation and completion.

Collaborative Task Completion and Safety Features

The latest update enables multiple robots of different types to work together and accomplish tasks. A video showcases how Apollo 2 instructs Google’s dual-arm robot to organize tools in a bin while cleaning a garage. Google DeepMind emphasizes that Gemini Robotics ER 2 is its safest model to date, with enhanced capabilities to detect human presence, trigger safety protocols, and halt robot movement if a person approaches too closely.

Improvements to Gemini Robotics On-Device Model

Additionally, Google DeepMind has made enhancements to the Gemini Robotics On-Device Model, allowing it to run locally on robots without an internet connection. This updated model can quickly adapt to new robot configurations, including those with varying shapes, sensors, and degrees of freedom.