🔍 Intelligence Synthesis JP/EN
Ethics · Regulation · Technology — Pharma Practice Notes

AI Tech Flash ── 2026-10-28

Google previews ER 2 reasoning model for robots

A robot-brain AI model is now in public preview on Google Cloud.

What changes

According to Google Cloud, enterprises can now pair a robot-brain AI model with its cloud platform. Simulation, evaluation and deployment can be handled in one environment.

In the company's announcement, a robot can check from video whether a step succeeded and redo only the failed step. It does not need to restart the whole workflow.

Why it matters

Google Cloud says physical operations require a robot to confirm each step succeeded before moving to the next. ER 2 is said to judge this by reasoning over continuous video.

The company says different types of robots can share an understanding of the workspace, divide labor and hand off tasks. That would allow work that a single machine could not finish alone.

What it is

Gemini Robotics Embodied Reasoning 2 (ER 2) is a specialized model built on Gemini. Google Cloud describes it as a robot's high-level brain.

Embodied reasoning means a robot understands situations and plans steps in the physical world. According to the company, ER 2 combines spatial and temporal video understanding, long-horizon planning and built-in safety guardrails.

Who and where

ER 2 is the latest embodied reasoning model from Google DeepMind. According to the Google Cloud blog, it is offered in public preview on the Gemini platform.

Users can start through Model Garden on that platform. The blog says enterprise identity and access management (IAM) and other security controls are available.

Why it was built

In Google Cloud's account, the problem being addressed is making robots adapt to real-world situations. Robots need to adjust to circumstances, correct mistakes and handle novel situations.

Running real robots demands safety and management tools, unlike chat-style AI use. The company says it therefore combines the DeepMind model with cloud operations features.

How it works

According to Google Cloud, developers declare low-level control interfaces as tools. Examples are Vision-Language-Action (VLA) models and navigation APIs. A VLA model turns images and language into robot actions.

Video, audio or text can be streamed straight into the model. ER 2 thinks about what comes next while it calls those tools to orchestrate actions.

The company says robots can be commanded in natural language and can explain their approach while acting. Operators can redirect them without technical language.

Key points

  1. According to Google Cloud, ER 2 is now in public preview on the Gemini platform.
  2. The company says it pinpoints key video frames with 91.3% accuracy.
  3. Google Cloud says simulation through live operation runs on one platform.

Conclusion

Google Cloud has begun offering DeepMind's robot reasoning model ER 2 in public preview. In the company's announcement, one model handles video understanding, planning and multi-robot coordination.

The figures and customer comments come from Google Cloud's own blog. Independent verification is not part of the cited source.

Quotes and sources

“Gemini Robotics ER 2 is now available in public preview on the Gemini platform.”
── Google Cloud, 2026-10-27
“Pinpoints the exact video frame where a critical event occurs (such as when to stop pouring a liquid) with 91.3% accuracy and a 0.96-second mean absolute frame distance”
── Google Cloud, 2026-10-27
“the ability to apply AI models to video opens a whole new world of possibilities.”
── Google Cloud, 2026-10-27

Source articles

Written by AI (Claude) from the cited articles, then checked by Jev (TypeSafe's judgement model), a Gemini fact-check and a verbatim check of every quote.