Google has unveiled Gemini Robotics 2, an AI model that DeepMind says moves robots beyond careful arm movements toward controlling an entire body at once.
According to Dataconomy, the model gives robots "full-body AI control" — meaning the system coordinates legs, torso and hands together rather than treating a gripper as an isolated tool. Robotics & Automation News reports that a humanoid built by Apptronik demonstrated this whole-body capability alongside the announcement.
The demos emphasize fiddly, everyday dexterity. MSN reports that the model enables humanoid robots to perform intricate tasks such as tying knots and collaborating on complex workflows, and quotes DeepMind CEO Demis Hassabis on the advance. The Independent describes humanoids screwing in light bulbs and tying bin bags, and notes that Gemini can now walk around inside a humanoid robot rather than staying on a screen.
Those sound like small chores, and that is the point. Screwing in a bulb requires holding your body steady while your fingers apply gentle, precise torque. Tying a bin bag means handling a floppy object whose shape changes as you touch it. Both have been notoriously hard for machines that excel at rigid, repeatable factory motions.
A dose of skepticism is warranted. TechRadar, writing about a robot that comically collapsed, argues that robot development is nowhere near where it needs to be — while allowing that a breakthrough from DeepMind could change that. Polished demo footage is not the same as reliability in a messy home.
Still, the direction matters: the same general-purpose AI models powering chatbots are now being wired directly into physical bodies. If that transfer holds up outside the lab, the gap between software intelligence and machines that can actually do your chores gets meaningfully smaller.