A granted patent is enforceable coverage — a claim a company can assert — so the useful question when one issues to a well-known robotics maker is concrete: what did the company just lock in, and where in its stack does that coverage sit? For Boston Dynamics, whose US12585292B2 issued under the March 24, 2026 grant date, the answer points away from the hardware the company is famous for. The grant is not about how a legged robot walks. It is about what the robot is sent out to do: inspect a location and notice when something has changed.

US12585292B2, titled "Location based change detection within image data by a mobile robot," describes instructing a robot to navigate to a location as part of an inspection mission, obtaining image data there, and detecting a change against a reference model to flag an anomaly. Its classification spread — G05D 1/689 and B62D 57/032 alongside several G06T image-processing classes (G06T 7/0002, G06T 7/70) — marks it as a perception-and-navigation filing, the software layer that turns a mobile platform into an inspection tool. The named inventors include Marco da Silva and Christopher Stathis, who recur across the company's recent issuances.

Based on the detected change and a reference model, the system can determine presence of an anomaly condition in the obtained image data.— Location based change detection within image data by a mobile robot, US12585292B2

A coverage block built around navigation and the inspection mission

The change-detection grant does not stand alone. Across recent grant cycles Boston Dynamics has issued a run of patents that fence the navigation-and-mapping layer an inspection robot relies on. US12508706B2 covers generating waypoints and edges and constraining a robot to stay within bounded regions while traversing toward a destination; US12461531B2 covers updating a waypoint-and-edge topological map with candidate alternate edges confirmed against sensor data. Both are classified in the G05D 1/02 navigation family. Read with the change-detection grant, they describe coverage over the full loop an inspection robot runs: build the map, route to the waypoint, capture the scene, compare it to a reference.

Two further grants fence the robot's response to the physical world it moves through. US12468300B2 covers detecting "negative obstacles" — missing terrain data large enough to swallow a foot — and classifying that area as a no-step region. US12449822B2 covers ground-clutter avoidance, using stored footfall locations from a prior run of a mission to identify obstacles and route the robot around them. These are the safety-and-reliability claims that determine whether a robot can be left to run an inspection route unattended — a different concern from the gait-and-balance patents that defined the company's earlier portfolio.

The contrast with that earlier work is itself a data point. Boston Dynamics' historical record — visible in the same assignee footprint — is dense with locomotion mechanics: gait controllers, step-path determination, slip detection, hydraulic-actuator valves and transmission overload protection. The recent grant cycles still touch hardware, but the weight has shifted. Several 2026 issuances read as control-and-perception software rather than mechanism: a screw actuator (US12564940B2) and sensor-noise mitigation for legged robots (US12570368B2) sit alongside the navigation and inspection grants, but the volume of recently issued claims clusters on the software that decides where the robot goes and what it records once it arrives. For a company whose public image is a backflipping machine, the issued coverage describes something more mundane and more commercial: a fleet of platforms that run repeatable routes and report what they see.

Pointing the sensor, and picking things up

The footprint also reaches into how the robot aims its instruments and handles objects. US12466075B2, classified in B25J 9/1697 and related manipulator classes, covers an autonomous-and-teleoperated workflow in which the robot determines an alignment pose, moves to it, and then points a sensor at a commanded location to capture data — the mechanics behind "go inspect that gauge." On the manipulation side, US12552030B2 covers grasping and placing multiple objects with a suction-based gripper by scoring candidate groups, and US12544932B2 covers modeling a loading-dock environment from camera images using a machine-learning model — coverage aimed squarely at warehouse and logistics work.

The classification pattern across the cluster reinforces the read. The change-detection, navigation and obstacle grants carry G05D 1/02-family codes for autonomous position-and-course control; the sensor-pointing and grasping grants carry B25J 9/16-family codes for programmed manipulator control; and several add G06T image-processing classes. What is comparatively thinner in this recent run is the B62D 57/032 legged-locomotion class that dominates the company's earlier issuances — it still appears, attached to the mobile-platform grants, but it is no longer the center of gravity. A grant cycle does not announce a strategy, but the CPC distribution is a fact: the recently issued claims sit on perception, navigation and manipulation classes more than on locomotion classes. The map-and-obstacle grants and the grasping grants together describe the bookends of an inspection-and-handling task: route there safely, then either record the scene or pick something up.

Stepping back, the shape of the recent block is the story. The change-detection grant that issued this cycle is the visible tip of a cluster whose center of gravity is the inspection-and-manipulation software — navigation maps, obstacle and no-step logic, sensor pointing, multi-object grasping, loading-dock perception — rather than the legged-locomotion mechanics the brand is associated with. For a reader tracking where an established robotics name is fencing its coverage, these grants indicate the company is investing in the layer that converts a mobile platform into a repeatable industrial-inspection and material-handling system. Each of the cited patents is now issued and assertable; collectively they describe enforceable coverage over how a robot is routed to a location, kept safe along the way, aimed at a target, and made to notice when the scene has changed.