Boston Dynamics adds Gemini AI to Spot

Mobile inspection robots are already useful. They walk routes, capture images, and keep people out of ugly places. But the hard part has always been the last mile — turning a pile of sensor data into something closer to judgment. That is what Boston Dynamics is trying to fix. The company has added Google DeepMind’s Gemini Robotics-ER 1.6 into its industrial inspection system for Spot, tied to the Orbit platform and its AIVI-Learning tools. The point is not to make Spot look smarter in a demo. The point is to make it better at the boring, expensive work of checking facilities for problems before humans have to step in. (bostondynamics.com) ### What actually changed? Boston Dynamics said it is integrating Gemini and Gemini Robotics-ER 1.6 into Orbit AIVI-Learning, the software layer that helps Spot understand what it sees during facility inspections. That means the news is bigger than “Spot got an AI upgrade.” It is really a workflow upgrade — Spot collects the data, Orbit organizes it, and Gemini helps interpret what matters inside a complex site. (bostondynamics.com) ### Why is instrument reading such a big deal? Because industrial inspections are full of analog weirdness. A gauge needle can sit at an odd angle. A sight glass can be cloudy. Lighting can be bad. Traditional automation is fine when the environment is tightly controlled, but plants are messy. DeepMind said instrument reading was a new capability unlocked in Gemini Robotics-(bostondynamics.com)re — practical maintenance checks, not flashy humanoid tricks. (deepmind.google) ### What does Gemini add that old vision systems did not? Basically, reasoning. Boston Dynamics says customers need more than object recognition. A normal computer-vision system might detect that a panel, valve, or meter exists. A stronger embodied reasoning model can help decide what the reading means, whether a result looks abnormal, and what should be checked next. DeepMind describes Gemini Robotics-ER (deepmind.google)tion, and retry decisions. (bostondynamics.com) ### Is this happening only in software? Mostly, yes — and that matters. Spot is already deployed as a mobile inspection robot. Boston Dynamics is not announcing a brand-new machine here. It is making the existing robot and software stack more useful, which is usually how industrial automation spreads in the real world. Companies buy upgrades that improve uptime and reduce manual rounds. They do not wait for science-fiction hardware. (bostondynamics.com) ### How does this fit with the bigger Boston Dynamics–DeepMind tie-up? It looks like the first concrete industrial payoff from a broader partnership. In January 2026, Boston Dynamics and Google DeepMind announced they would work together on bringing Gemini Robotics models to Atlas, the humanoid robot. The Spot inspection rollout shows the partnership is not just about future humanoids — it is already landing in current products where customers can use it now. (bostondynamics.com) ### Why start with inspections? Because inspections are a sweet spot for autonomy. The routes are repeatable, the value is easy to measure, and the shortage of skilled maintenance labor is real. If a robot can reliably notice a bad gauge reading, a leak, or a visual anomaly before a shutdown, the economics get compelling fast. That is a much easier sell than asking a robot to do fully general factory work on day one. (bostondynamics.com) ### So what is the catch? The catch is that “reasoning” in a plant has to be trustworthy, not just impressive. A system that occasionally misreads a gauge or overconfidently flags the wrong anomaly creates more work, not less. Boston Dynamics is pitching a more sophisticated inspection layer, but the real test will be whether customers trust those calls enough to fold them into maintenance routines. That part takes time. (bostondynamics.com) ### Bottom line? This is a practical AI story disguised as a robot story. Boston Dynamics is using Gemini to push Spot from route-following data collector toward something more like a junior inspector — still narrow, still supervised, but a lot closer to useful judgment in the field. (bostondynamics.com)