A granted patent is not a research note; it is enforceable coverage, a claim a company can assert. So when a single U.S. grant cycle issues ten patents to one industrial-robot maker on the same day, the question for a business reader is concrete: what did the company just lock in, and where in the stack does that coverage sit? For FANUC, whose patents issued under the May 19, 2026 grant date, the answer is unusually legible. The ten grants do not concentrate on the visible hardware — the arm, the joints, the end-effector. They concentrate on the controller: the software and safety logic that sits between a human operator and the machine.

The clearest example is US12632028B2, a numerical control device that switches a robot's payload setting and feeds it into the torque math. The grant describes selecting a payload configuration, transmitting it to the robot control device, and applying it to the inverse-dynamics torque calculation — coverage over the handoff between a machine tool's numerical controller and the robot working alongside it. Its classification spread (B25J 9/1607, B25J 9/1633, B25J 9/1638 alongside G05B numerical-control classes) marks it as a control-integration filing, the kind that matters on a factory line where a robot and a CNC machine share a cell.

By allowing payload settings to be switched and reflected in torque computation, the device improves efficiency and accuracy of robot control under varying payload conditions.— Numerical control device and numerical control system, US12632028B2

A coverage block built around safety and authorization

Two of the week's grants are explicitly about who is allowed to change a robot's behavior, and how that change is verified. US12629832B2, a robot control device, covers a workflow in which a setting parameter is applied to a safety parameter "for verification in a verification mode," then copied and re-applied to a second safety parameter after the system switches to a safety mode — a two-stage path designed so that safety configuration is validated before it goes live. Its single classification, B25J 9/1666, places it squarely in robot safety control. Alongside it, US12629835B2 covers a management device that checks whether a modification to a robot's operation program has been authorized before that program runs, and executes an "accident prevention process" if it has not. Read together, the two grants give FANUC issued coverage over the governance layer of an industrial robot: not what the arm does, but how a change to what the arm does is permitted and checked.

That governance theme extends into the teaching grants. US12629824B2, a teaching device and robot system, covers an interface that displays an execution-history screen, lets an operator select a single past run, restores it to a program, and modifies its configuration parameters — coverage over the human-in-the-loop tooling that programs a robot on the shop floor (classes B25J 9/163, B25J 9/1697, B25J 13/06). US12629771B2 extends the same teaching idea into a laser-machining context, covering a system that generates teaching data by iteratively adjusting a beam angle against reflected-light intensity. These are not glamorous claims, but they fence the interface through which a robot is actually configured for a task — the surface a customer touches.

Learning, scheduling and the parts that pencil out

The cycle also includes a grant on how a robot acquires a skill in the first place. US12629837B2, a method for robot assembly-skill learning, covers an actor-critic reinforcement-learning controller coupled to a compliance controller for high-precision assembly: an actor network supplies target-position adjustments based on robot state feedback, and a critic network trains the actor against reward data, after which the critic is dropped. The grant is classified in B25J 9/1687 and B25J 9/163, the learning-and-control corner of the manipulator class. It gives FANUC issued coverage over a specific machine-learning architecture for the kind of force-sensitive assembly task — inserting, mating, fitting parts — that has historically resisted automation. A separate machine-learning grant, US12632031B2, covers a learned model that predicts cleaning conditions for the inside of a machine tool, extending the learning theme into maintenance.

Rounding out the block are grants closer to throughput economics. US12632038B2, a control device, covers analyzing a control program that runs multiple controlled objects in parallel, detecting "wasted time" from waiting between them, and adjusting axis speed and acceleration to reduce it — coverage over cycle-time optimization, which is the metric that determines whether an automation cell pays for itself. US12632025B2 covers a display that distinguishes actual-cut from air-cut portions of an oscillation-cutting tool path, an operator-visibility filing in the same numerical-control family.

Set against FANUC's position as one of the volume leaders in the week's robotics-tagged grants, the shape of the block is the story. The coverage FANUC issued this cycle runs through the controller, the safety logic, the teaching interface and the learning algorithm — the parts of an industrial robot that are software and governance rather than steel. That places the company's freshly granted footprint on the layer where an incumbent's installed base, control architecture and operator workflows are hardest for a rival to replicate, and where switching costs for a factory already running FANUC controllers are highest. None of these grants describes a new robot; collectively they describe enforceable coverage over how a robot is configured, secured, taught and optimized. For a reader tracking where an automation incumbent is fencing, the controller layer — not the arm — is where this week's claims landed, and each of the ten is now issued and assertable rather than merely pending.