How is Gemini Robotics ER 2 different from Gemini Robotics 2?

Gemini Robotics 2 is a Vision-Language-Action (VLA) model that directly moves a humanoid's joints. Gemini Robotics ER 2 sits above it as the embodied-reasoning brain: it plans the multi-step task, watches video for progress, decides which VLA or API to call, and coordinates multiple robots in a shared scene. The two are designed to be paired.

What does the video understanding actually enable?

Gemini Robotics ER 2 ingests a live video stream and tracks task progress across five discrete levels, hitting 57.4% classification accuracy. It also finds critical moments in the stream with 91.3% accuracy and a mean absolute distance of 0.96 seconds, so the robot can catch a spill, notice a misplaced part, or self-correct mid-task instead of blindly running the plan.

How do I try Gemini Robotics ER 2 today?

Gemini Robotics ER 2 is exposed through the Gemini API and Google AI Studio, with a robotics-overview page and a Python samples repo at github.com/google-gemini/robotics-samples. Google DeepMind pairs it with the underlying Gemini Robotics 2 VLA for full-body humanoid control shown in the accompanying blog post.

How does it handle multiple robots?

Gemini Robotics ER 2 lets diverse robots — humanoids, arms, mobile platforms — share a semantic map of the scene and reason jointly about who does what. DeepMind demos two different robots collaborating on the same task without hardcoded coordination rules, using ER 2 as the shared planner.

Google DeepMind · 2026-07-30 · major

Gemini Robotics ER 2 — the planning brain that watches video and coordinates robots

Gemini Robotics ER 2 handles high-level reasoning for robots: continuous video, multi-step task planning, tool orchestration, and multi-robot coordination. Hits 91.3% on moment-finding and 57.4% on progress classification.

Gemini Robotics ER 2 cover graphic on the Google blog

DeepMind's embodied-reasoning brain plans multi-step robot work, watches the video, and orchestrates multiple robots at once.

Quick facts

Maker	Google DeepMind
Role	High-level planning brain paired with Gemini Robotics 2 VLA
Inputs	Continuous video + language + robot state
New capabilities	Real-time video understanding, multi-robot coordination, tool orchestration
Progress classification	57.4% accuracy across 5 progress levels
Moment-finding	91.3% accuracy, 0.96s mean absolute distance
Availability	Gemini API + Google AI Studio, sample code on GitHub

What is it?

Gemini Robotics ER 2 is Google DeepMind's high-level embodied-reasoning model — the successor to Gemini Robotics-ER 1.6. It chats with humans, understands the physical world, plans multi-step tasks, and orchestrates other tools (including Vision-Language-Action models and APIs) so a robot can carry the plan out.

How does it work?

The ER 2 model ingests a continuous video feed alongside language and robot state, then tracks progress across five discrete levels and flags critical moments in the stream (57.4% and 91.3% accuracy respectively in DeepMind's tests). It calls out to a paired VLA — such as Gemini Robotics 2 — to execute low-level motion, and shares a semantic scene understanding so multiple heterogeneous robots can coordinate on the same task.

Why does it matter?

Splitting the robot stack into a Gemini Robotics 2 VLA for motion and a Gemini Robotics ER 2 brain for reasoning lets developers upgrade one without retraining the other. The step-change in video understanding also means robots can self-correct mid-task instead of failing quietly, and multi-robot support opens the door to warehouse and home scenarios that need more than one machine cooperating.

Who is it for?

robotics developers building agent stacks on Gemini

Frequently asked questions

How is Gemini Robotics ER 2 different from Gemini Robotics 2?: Gemini Robotics 2 is a Vision-Language-Action (VLA) model that directly moves a humanoid's joints. Gemini Robotics ER 2 sits above it as the embodied-reasoning brain: it plans the multi-step task, watches video for progress, decides which VLA or API to call, and coordinates multiple robots in a shared scene. The two are designed to be paired.
What does the video understanding actually enable?: Gemini Robotics ER 2 ingests a live video stream and tracks task progress across five discrete levels, hitting 57.4% classification accuracy. It also finds critical moments in the stream with 91.3% accuracy and a mean absolute distance of 0.96 seconds, so the robot can catch a spill, notice a misplaced part, or self-correct mid-task instead of blindly running the plan.
How do I try Gemini Robotics ER 2 today?: Gemini Robotics ER 2 is exposed through the Gemini API and Google AI Studio, with a robotics-overview page and a Python samples repo at github.com/google-gemini/robotics-samples. Google DeepMind pairs it with the underlying Gemini Robotics 2 VLA for full-body humanoid control shown in the accompanying blog post.
How does it handle multiple robots?: Gemini Robotics ER 2 lets diverse robots — humanoids, arms, mobile platforms — share a semantic map of the scene and reason jointly about who does what. DeepMind demos two different robots collaborating on the same task without hardcoded coordination rules, using ER 2 as the shared planner.

Try it

ai.google.dev/gemini-api/docs/robotics-overview