Feet to Fingertips: Gemini Robotics 2 Ships Whole-Body Control and Open ER 2

Google DeepMind on July 30, 2026 released Gemini Robotics 2 — not a single model, but a three-layer stack meant to turn multimodal AI into whole-body robot control, dexterous manipulation, and multi-robot teamwork.

The launch arrives as U.S. policy tightens access to foreign humanoid hardware, making software portability across embodiments more strategically valuable than another single-vendor demo.

Feet to fingertips: whole-body humanoids

Previous Gemini Robotics releases focused on upper-body tabletop tasks. Gemini Robotics 2, the vision-language-action (VLA) model, now controls full humanoids — walking, crouching, reaching, and placing objects in cluttered spaces.

Published success rates add nuance: Apollo 2 with Inspire hands picked from shelves 76.3% of the time; floor picks were 45.7%. Five-finger SharpaWave tasks remain challenging — screw-in success around 36%.

ER 2: the reasoning brain goes public

Gemini Robotics ER 2 is available now via the Gemini API and Google AI Studio. It plans multi-step tasks from live video, tracks progress, and coordinates unlike robots — including demos where a wheeled base and a humanoid split labor.

On-device adaptation in hours

Gemini Robotics On-Device 2 targets deployments that cannot tolerate cloud latency. DeepMind claims adaptation to new bi-arm embodiments in a few hours with fewer than 200 examples.

The VLA and on-device models go to early-access partners — Apptronik, Boston Dynamics, Agile Robots.

Safety as a benchmark

DeepMind introduced ASIMOV-Agentic to measure whether reasoning agents refuse unsafe tool calls and request human help when uncertain.

### Sources

• Google DeepMind — Gemini Robotics 2 brings whole body intelligence to robots (July 30, 2026)
• Google — Introducing Gemini Robotics ER 2 (July 30, 2026)
• Unite.AI — Google Ships Gemini Robotics ER 2 With Multi-Robot Teamwork (July 30, 2026)
• Axios — Google updates its AI software for robots (July 30, 2026)