Every morning on factory floors, a robot now scans the needle position on a pressure gauge, then interprets the flow meter digits buried inside a tangle of pipes. The task that once required a human to walk the line and squint at analog dials is being handed over to machine vision. This week, Google released the model that makes that scene possible: Gemini Robotics-ER 1.6.
Spatial Reasoning and Instrument Reading, Two Core Improvements
Google announced Gemini Robotics-ER 1.6, a model designed to strengthen what the company calls "embodied reasoning" — the ability for robots to understand the physical world. Compared to its predecessor, Gemini Robotics-ER 1.5, and to Gemini 3.0 Flash, the new model improves in three areas: visual and spatial understanding, task planning and success detection, and a newly added instrument reading capability.
The instrument reading feature emerged from a collaboration with Boston Dynamics. It enables robots to read complex gauges and sight glasses — the transparent windows that show internal liquid levels — a task that was previously nearly impossible for robotic systems. This opens up precision measurement work in environments where analog instruments remain the standard.
Developers can access the model starting today through the Gemini API and Google AI Studio. Google provides a Colab notebook to help get started, which includes model setup and prompt examples for embodied reasoning tasks.
What Used to Require a Human Holding a Gauge
Traditional robot systems operated at the level of following commands. They needed explicit instructions like "rotate 30 degrees to the right" and could not adapt to changes in their environment on their own. Gemini Robotics-ER 1.6 changes that. The model acts as a high-level reasoning layer: it can call tools like Google Search directly to find information, execute vision-language-action models (VLAs, which generate physical robot actions from images and language commands), or run custom functions.
When given a command like "check if that pressure gauge is in the normal range," the model first locates the gauge, reads the needle angle, compares it against the normal range, and returns a result. The entire process runs as a single reasoning chain.
For developers, the most immediate change is in how tools connect. Gemini Robotics-ER 1.6 natively supports calling third-party functions, reducing the need for custom middleware. Here is the core setup code from Google's Colab example:
import google.generativeai as genaimodel = genai.GenerativeModel('gemini-robotics-er-1.6')
response = model.generate_content(
"이 압력 게이지의 현재 값을 읽고 정상 여부를 판단해줘",
safety_settings={
"HARM_CATEGORY_DANGEROUS_CONTENT": "BLOCK_NONE"
}
)
Robot Reasoning Enters the Industrial Floor
As the Boston Dynamics collaboration shows, this model is designed to move beyond the lab and into factories and logistics centers. Instrument reading is essential in chemical plants, semiconductor manufacturing, and energy facilities. Previously, workers had to walk regular rounds to check gauges. Now a robot can take over that task.
The improvements in spatial reasoning also allow robots to point at objects, count items, and determine whether a task succeeded — all in complex environments. That translates directly into better autonomous logistics robots and warehouse management systems.
Developers can integrate the model into existing robot control pipelines through the Gemini API. Testing is free in Google AI Studio, and production use is billed by API consumption. Google has not yet published exact pricing, but it is expected to be similar to Gemini 3.0 Flash.
The shift is subtle but real: robots are moving from machines that move to agents that understand and judge their environment.
