Google DeepMind Launches Gemini Robotics 2 for Full-Body Humanoid Control

One-line summary: Google DeepMind's Gemini Robotics 2 series gives humanoid robots full-body autonomy, multi-step reasoning, and on-device adaptability — raising the bar for physical AI at every level of the stack.

Key Points

• Gemini Robotics 2 is a VLA (vision-language-action) model translating vision and language into robot actions; unlike its predecessor, it controls the entire body, not just the upper half
• Three variants: Gemini Robotics 2 (full-body VLA), Gemini Robotics ER 2 (embodied reasoning + multi-robot coordination), Gemini Robotics On-Device 2 (locally deployable, adapts to new robot forms with only hours of training data)
• Demonstrated on Apptronik's Apollo 2 humanoid performing walking, crouching, and object manipulation in real time
• All three models now available in early access

Why It Matters

Physical AI is the next frontier after language models. By packaging whole-body control, task reasoning, and edge inference into a single model family, Google is positioning itself as the operating system for the humanoid robotics industry — a market heating up with Figure, Agility Robotics, and others already public.

Read More

• Gemini Robotics 2 brings whole body intelligence to robots — Google DeepMind Blog
• Google DeepMind debuts Gemini Robotics 2 — SiliconANGLE