Boston Dynamics adds Gemini to Spot

Industrial inspection robots are useful for one reason — they can go places people would rather not keep checking by hand. But the hard part has never been walking. Spot already knew how to walk. The hard part was understanding messy, variable scenes well enough to do real inspection work without a human constantly narrowing the task. That is what Boston Dynamics is trying to fix by plugging Google DeepMind’s Gemini Robotics-ER 1.6 into Spot and its Orbit software stack. (bostondynamics.com) ### What actually changed? Boston Dynamics said Orbit AIVI-Learning now uses Gemini and Gemini Robotics-ER 1.6, with the upgrade aimed squarely at industrial inspections. This is not a vague “AI-powered” label. The company is tying the model to specific jobs inside facilities — reading gauges, estimating sight-glass fullness, counting pallets, spotting puddle(bostondynamics.com)ys the Gemini-powered version is already live for enrolled customers, with the transition effective April 8, 2026. (bostondynamics.com) ### Why is gauge reading a big deal? Because gauges are the annoying real-world test. A scripted robot can revisit the same checkpoint all day, but analog dials, reflections, odd camera angles, and cluttered backgrounds make “just read the instrument” much harder than it sounds. DeepMind built instrument reading into Gemini Robotics-ER 1.6 after working with (bostondynamics.com)mers were getting stuck. (deepmind.google) ### What does Gemini add that Spot lacked? Basically, reasoning. Spot was already a strong mobile platform, but most industrial autonomy has been brittle — predefined routes, predefined checks, predefined thresholds. Gemini Robotics-ER 1.6 is meant to let the robot interpret a scene, decide what matters, and call other tools when needed. DeepMind says the model improves spatial reasoning, m(deepmind.google)amics is also surfacing “transparent reasoning,” so operators can inspect why the system reached a conclusion instead of treating the model like a black box. (deepmind.google) ### Why does Orbit matter here? Because Spot is only half the story. Orbit is the layer that turns robot patrols into a usable inspection system — scheduling routes, collecting images, comparing changes over time, and pushing results to operations teams. Adding Gemini at the Orbit AIVI-Learning layer means Boston Dynamics can improve inspection performance from the cloud and roll out new vi(deepmind.google) explicitly pitching this as a zero-downtime upgrade path. (bostondynamics.com) ### Is this just a demo? Not really — and that is the interesting part. Boston Dynamics says several thousand Spot robots are already deployed, and IEEE Spectrum noted inspection is one of the few legged-robot jobs that has proven commercially viable at real scale. So this announcement lands differently from a lab video of a robot doing chores. It is aimed at(bostondynamics.com)ics are easier to justify. (spectrum.ieee.org) ### What is the catch? Reliability. Industrial customers do not care that a robot can sometimes read a gauge. They care whether it can do it every day, in bad lighting, with odd mounting positions, and without flooding teams with false alarms. Boston Dynamics says accuracy improves on key inspections, including digital display readings, but it has not published a full customer-g(spectrum.ieee.org)ng-run uptime and error rates in the field. (bostondynamics.com) ### Why does this matter beyond Spot? Because this is what embodied AI looks like when it leaves the benchmark chart and gets attached to a job budget. The near-term win is boring in the best way — fewer manual rounds, faster hazard detection, and better coverage of repetitive checks. The bigger shift is that robots are starting to move from “follow the route” to “understand the scene.” If that works reliably, inspection is just the first foothold. (spectrum.ieee.org)