One Brain, Any Body: Carolina Parada on Gemini Robotics 2 and the Physical-AI Layer

Google DeepMind did not hold a press conference on July 30, 2026. What it published was a milestone blog post and a developer rollout of Gemini Robotics ER 2 — and in coverage that followed, Carolina Parada, who leads robotics at DeepMind, offered the clearest public articulation yet of the lab's physical-AI strategy. This piece reconstructs her statements from DeepMind's blog, Google's product announcement, and verified reporting. It is not a private interview.

"Bring AI into the physical world"

In reporting from The Next Web, Parada framed the release in terms of universality: "Our goal is to bring AI into the physical world and then build the intelligence layer that can be used by every robot." The July 30 announcement is not one humanoid demo; it is a stack meant to port across embodiments.

DeepMind's blog, attributed to Parada, describes Gemini Robotics 2 as taking "its first literal steps" toward intelligent whole-body control, advanced dexterity, and multi-robot collaboration. The new vision-language-action (VLA) model extends control from feet to fingertips.

Three models, one architecture story

• Gemini Robotics 2 — VLA motor control for humanoids and bi-arm platforms.
• Gemini Robotics ER 2 — embodied reasoning that plans multi-step tasks and orchestrates multiple robots.
• Gemini Robotics On-Device 2 — local VLA adapting to new bi-arm embodiments in a few hours with fewer than 200 examples.

Only ER 2 is broadly available via the Gemini API and Google AI Studio on launch day.

Dexterity with published success rates

On Apptronik's Apollo 2 with Inspire hands, shelf picks succeeded 76.3% of the time; floor picks were 45.7%. Five-finger SharpaWave manipulation remains explicitly challenging, with screw-in tasks at 36% success.

Multi-robot collaboration

ER 2 can assign subtasks across unlike machines and steer Boston Dynamics Spot on spoken command. DeepMind introduced ASIMOV-Agentic to test whether reasoning agents refuse unsafe tool calls.

Milestone, not finish line

Parada's team is careful: Gemini Robotics 2 makes robots more general, but humans still outclass the best tidying demo. ER 2 is live for developers; the VLA path remains gated to early-access partners.

### Sources

• Google DeepMind — Gemini Robotics 2 brings whole body intelligence to robots (July 30, 2026)
• Google — Introducing Gemini Robotics ER 2 (July 30, 2026)
• The Next Web — Google DeepMind's Gemini Robotics 2 controls whole humanoids (July 30, 2026)
• Axios — Google updates its AI software for robots (July 30, 2026)