Google rolls out robotics 1.6

Google DeepMind has released Gemini Robotics‑ER 1.6, a new model for industrial robots that can read gauges and plan physical tasks. (deepmind.google) Robots in factories and utility sites need a “reasoning” layer that turns camera views and language prompts into decisions about where to look, what to inspect, and whether a job is finished. Google said Robotics‑ER 1.6 improves spatial reasoning, multi‑view understanding, task planning, and success detection. (deepmind.google) Google said the new model adds instrument reading, including complex gauges and sight glasses, after work with Boston Dynamics. The company published the update on April 14, 2026, and said developers can access it through the Gemini API and Google AI Studio. (blog.google) This is the second year of Google’s push to split robot intelligence into parts: one model that reasons about the scene and another that handles motion. When Google introduced Gemini Robotics and Gemini Robotics‑ER in March 2025, it described ER as the system that lets roboticists run their own programs using embodied reasoning. (blog.google) The practical target is industrial inspection, where workers still spend hours checking dials, fluid levels, debris, and equipment status across large sites. Boston Dynamics said on April 14 that it is integrating Gemini and Gemini Robotics‑ER 1.6 into Orbit AIVI‑Learning for Spot deployments in facilities. (therobotreport.com) Boston Dynamics said the system is aimed at safety checks, asset monitoring, and materials tracking, including conveyor damage, sight glass levels, pallet counts, and standing liquid. It also said AIVI‑Learning now supports gauges, a direct fit with Google’s new instrument‑reading feature. (therobotreport.com) Google’s developer documentation describes Robotics‑ER 1.6 as a vision‑language model in preview, not a full robot controller. The model can identify objects, return 2D points in an image, break natural‑language commands into subtasks, and pass those outputs to robot control software or other tools. (ai.google.dev) Google also says the model can call tools such as Google Search, vision‑language‑action models, or third‑party functions while carrying out a task plan. In Google’s setup, that makes Robotics‑ER 1.6 the planner for a robot rather than the motor system itself. (deepmind.google) The company paired the launch with a safety claim, calling Robotics‑ER 1.6 its safest robotics model so far on adversarial spatial‑reasoning tests. Its developer page also warns that generative models can make mistakes and that operators remain responsible for keeping robots in safe environments. (blog.google) (ai.google.dev) For now, Google is selling the software layer and Boston Dynamics is tying it to inspection hardware already used in plants and warehouses. The immediate use case is not a general household robot; it is a machine that can walk a site, read a dial, and decide what needs attention next. (deepmind.google) (therobotreport.com)