Google DeepMind on Thursday debuted Gemini Robotics 2, a family of AI models that the company says can "control entire humanoid robots" with whole-body motions from feet to fingertips3,4. The release marks a significant expansion from the prior version, which focused on upper-body control.
The model series comprises a trio of new sub-models, one of which is publicly available to developers as of the announcement. Among them is Gemini Robotics ER 2, an upgraded "embodied reasoning" model that Google DeepMind claims represents a significant leap over the previous 1.6 release. Gemini Robotics ER 2 is integrated with the Gemini Live API, giving developers access to the model's capabilities.
Google DeepMind says Gemini Robotics 2 enables robots to walk, crouch, stretch, and manipulate objects, with improved dexterity including for machines with complex humanoid hands. The company also says the model can continuously analyze changing environments and accomplish more complex tasks than its predecessor. According to Google DeepMind, the system can automate chores comprising hundreds of steps2.
A key capability highlighted in the announcement is multi-robot collaboration: Google DeepMind says the Gemini Robotics 2 series enables multiple autonomous machines to collaborate on a single task. Videos shared by Google show Apptronik's Apollo 2 robot bending over to pick up a watering can and finding and retrieving specific items from a shelf.
Google DeepMind frames the work under the banner of "physical AGI" — the goal of creating a generalist robot that can perform any task a human could when given a natural-language instruction1. Wired noted that deploying AI into the physical world carries risks alongside the capability gains.
The announcement lands the same day Google Cloud expanded access to its Gemini Enterprise Agent Platform, which reached general availability with infrastructure for building, securing, and governing AI agents at enterprise scale ctx. ANALYSIS The pairing of a robotics-focused model family with a broadened enterprise agent platform places the Gemini line across both digital and physical automation domains on the same news cycle.
The shift from upper-body-only control to whole-body humanoid operation, combined with multi-robot coordination and multi-hundred-step task automation, represents a qualitative change in the scope of tasks being targeted with these robotics models. Whether third-party hardware partners beyond Apptronik adopt the platform at scale will be a key variable in determining real-world impact.