What changed
Google DeepMind describes Gemini Robotics 2 as a VLA model that converts vision and language input into motor control. The same release describes whole-body humanoid control, advanced dexterity and multi-robot collaboration.
Why it matters for embodiment
The significance is architectural: intelligence is presented as a layer that can adapt across different physical platforms rather than being tied to one robot body.
Related layer
Gemini Robotics ER 2 is described as the higher-level embodied reasoning model, while On-Device 2 targets local execution on robotic hardware.