DeepMind’s robotics model 1.6

Robots have to connect what they see to what they do, and Google DeepMind says its new Gemini Robotics-ER 1.6 is built to improve that link. The model was released on April 14 and is aimed at physical tasks such as planning, checking whether a job is finished, and reading instruments. (deepmind.google) DeepMind describes the system as a high-level reasoning model for robots, not the motor controller itself. It can call tools, including Google Search, vision-language-action models, and other user-defined functions, through the Gemini application programming interface and Google Artificial Intelligence Studio. (deepmind.google) The update focuses on three concrete skills: spatial reasoning, multi-view understanding, and instrument reading. DeepMind said version 1.6 improved over Gemini Robotics-ER 1.5 and Gemini 3.0 Flash on tasks such as pointing, counting, and success detection. (deepmind.google) Spatial reasoning here means a robot can identify where objects are, compare them, and map a movement from one place to another. DeepMind said the model can use pointing as an intermediate step for harder jobs, such as finding every object small enough to fit inside a specific cup. (deepmind.google) Multi-view understanding means the robot can judge a task using more than one camera angle instead of a single image. DeepMind said that helps with success detection, which is the basic question every robot has to answer after acting: did the task actually get done. (deepmind.google) The new piece is instrument reading, which lets robots read gauges and sight glasses in industrial settings. DeepMind said that use case came from work with Boston Dynamics, whose Spot robot is already used for inspection rounds in factories, plants, and other facilities. (deepmind.google (bostondynamics.com) Boston Dynamics said on April 15 that it is integrating Gemini and Gemini Robotics-ER 1.6 into Orbit AIVI-Learning, its cloud-hosted visual inspection system for Spot. The company said the update adds support for gauges and expands tasks such as sight-glass measurement, pallet counting, and detection of standing liquid. (bostondynamics.com) That matters for the kind of work Spot already does: repeated inspection in places with hazardous machinery, hard-to-reach equipment, and around-the-clock monitoring needs. Boston Dynamics says Spot can collect visual, thermal, and acoustic data, while Orbit is the software layer that organizes alerts, trends, and fleet management. (bostondynamics.com (bostondynamics.com) Boston Dynamics has also been testing Gemini-style reasoning outside fixed inspection routes. In a separate post published April 15, the company said a 2025 hackathon used Gemini Robotics-ER 1.5 with Spot in a home setting, where engineers relied on natural-language prompts and lightweight tool scripts instead of writing a rigid step-by-step state machine. (bostondynamics.com) The companies had already widened their relationship earlier this year. At Consumer Electronics Show on January 5, Boston Dynamics and Google DeepMind said they would integrate Gemini Robotics models with Atlas, Boston Dynamics’ humanoid robot, for industrial tasks starting with automotive work. (bostondynamics.com) For now, the immediate release is software, not a new robot. DeepMind is putting Gemini Robotics-ER 1.6 in developers’ hands through the Gemini application programming interface and Google Artificial Intelligence Studio, while Boston Dynamics is plugging the same reasoning layer into Spot’s inspection stack. (deepmind.google)