AI อะไรเนี่ย
Google / DeepMind

Gemini Robotics ER 1.6: Enhanced Embodied Reasoning for Real-World Robot Tasks

Model

Gemini Robotics ER 1.6: Enhanced Embodied Reasoning for Real-World Robot Tasks

TL;DR

  • Google DeepMind released Gemini Robotics-ER 1.6, a major upgrade to its reasoning-first model for robots.
  • The model improves embodied reasoning so robots can perceive, plan, and act in physical environments more reliably.
  • It adds a new instrument reading capability — robots can now read pressure gauges and sight glasses, developed with Boston Dynamics.
  • Available today via the Gemini API and Google AI Studio with a developer Colab.

For robots to be genuinely useful in factories, labs, and homes, they need to do more than execute pre-programmed steps — they have to reason about the physical world they share with us. That capability, often called "embodied reasoning," is what bridges digital intelligence with real-world action. With Gemini Robotics-ER 1.6, Google DeepMind is pushing that bridge much further by giving developers a high-level reasoning model purpose-built for physical agents.

Compared to the previous Gemini Robotics-ER 1.5 and Gemini 3.0 Flash, the new model shows clear gains on the foundational skills that make a robot trustworthy in the real world: more accurate pointing at objects in cluttered scenes, more reliable counting, and stronger success detection so the robot can confirm when a task is actually finished. These are the kinds of unglamorous capabilities that decide whether a robot is a demo or a deployable system.

The headline new capability is instrument reading, developed in close collaboration with Boston Dynamics. Robots powered by ER 1.6 can interpret complex pressure gauges and sight glasses — exactly the sort of routine inspection work that drains human operators in industrial settings. Instead of replacing the inspector, a robot can now do the rounds and flag the readings that need attention, freeing skilled workers for the judgment calls that still need a human.

Architecturally, Gemini Robotics-ER 1.6 is positioned as a high-level reasoning model rather than a monolithic controller. It natively calls external tools — including Google Search for fresh information, dedicated Vision-Language-Action (VLA) models for low-level motor control, and any developer-defined function. That separation keeps the reasoning layer flexible while letting teams plug in the specialized policies they already have for grasping, navigation, or manipulation.

For builders, the access story is simple: the model is available today through the Gemini API and Google AI Studio, and DeepMind ships a developer Colab with example prompts and configurations for embodied reasoning tasks. If you have been waiting for a reasoning layer that can sit on top of your existing robotics stack without forcing a rewrite, ER 1.6 is the most pragmatic option Google has shipped so far.

Summary

  • Gemini Robotics-ER 1.6 is Google DeepMind's latest reasoning-first model for real-world robotics tasks.
  • It improves spatial reasoning, task planning, and success detection over ER 1.5 and Gemini 3.0 Flash.
  • New instrument reading support (built with Boston Dynamics) handles pressure gauges and sight glasses.
  • Developers can start today via the Gemini API and Google AI Studio with the provided Colab.

Source: Gemini Robotics ER 1.6 Announcement