Google DeepMind gives Gemini a bigger role in physical robots

Google DeepMind introduced Gemini Robotics 2, a robotics model aimed at bringing Gemini-style reasoning into machines that have to move, grasp and coordinate in the physical world.

The company says the model is designed for whole-body control, not just isolated arm movements. In practice, that means a robot can use visual understanding and task reasoning to handle actions that involve posture, balance, hands and surrounding objects together.

Google also described Gemini Robotics-ER 2, an embodied-reasoning model focused on video understanding, task orchestration and multi-robot collaboration. That second model is meant to help robotics systems interpret what is happening around them, break work into steps and coordinate tools or other robots.

The important part is not that robots suddenly become general-purpose household helpers. The launch is still framed as research and robotics development. The practical signal is that Google is trying to move Gemini beyond screens: from answering, coding and analyzing media into systems that can plan and act through physical hardware.

For readers watching AI strategy, this matters because robotics is one of the clearest paths from digital assistants to automation in factories, labs, warehouses and eventually homes. Better video understanding and orchestration do not solve robot hardware on their own, but they make the software layer more capable and easier to adapt across different robot bodies.