Spot accepts plain‑English

Boston Dynamics’ Spot can now take plain-English instructions for inspection work through Google DeepMind’s Gemini Robotics-ER, with the feature now in preview. (deepmind.google) Google DeepMind said on April 14 that Gemini Robotics-ER 1.6 is available through the Gemini API and Google AI Studio, replacing the earlier `gemini-robotics-er-1.5-preview` model name with `gemini-robotics-er-1.6-preview`. (ai.google.dev) Spot is Boston Dynamics’ four-legged robot for industrial sites, and the company said its new Orbit AIVI-Learning integration uses Gemini and Gemini Robotics-ER 1.6 to help the robot inspect facilities and answer natural-language requests. (bostondynamics.com) Embodied reasoning is the part of robotics that lets a machine connect what it sees to what it should do next, like planning a route, checking whether a valve is open, or deciding if a task is finished. Google DeepMind says Gemini Robotics-ER 1.6 is built for spatial logic, task planning, and success detection. (deepmind.google) Google said the new release adds instrument reading, which lets robots read gauges and sight glasses, and said that use case came from work with Boston Dynamics. The company also said the model improves spatial and physical reasoning over the prior version. (deepmind.google; ai.google.dev) Boston Dynamics said the system checks task completion across multiple camera views instead of relying on a single image, a setup aimed at helping Spot confirm whether an inspection step actually succeeded. (bostondynamics.com) This builds on Google DeepMind’s March 2025 launch of Gemini Robotics and Gemini Robotics-ER, two Gemini 2.0-based models for robots. In that rollout, Google described Gemini Robotics as the model that can directly control robots, while Gemini Robotics-ER was designed to support roboticists’ own programs with spatial understanding and reasoning. (blog.google) Boston Dynamics said an earlier Spot demo used Gemini Robotics-ER 1.5 and grew out of a 2025 internal hackathon, where the company tested ways for Spot to understand rooms, objects, and spoken requests beyond a standard preplanned Autowalk mission. (bostondynamics.com) For customers, the shift is less about teaching Spot new tricks than changing how operators assign work: typed or spoken instructions can now describe the task in everyday language, while the model handles planning, visual checks, and instrument reading in the background. (ai.google.dev; bostondynamics.com) The feature is still in preview, so the immediate story is not a mass rollout but a new layer of software on top of an existing industrial robot. Spot already walked facilities; now Boston Dynamics and Google are trying to make it understand inspection jobs the way a human supervisor describes them. (ai.google.dev; deepmind.google)