One granted patent tells you what a company built and protected. A run of them, landing across every part of the same machine, tells you where the company is putting its weight. On June 30, 2026, Figure AI added a grant covering a humanoid robot's head-and-neck assembly — US12667974B2, directed to a two-shell head housing that conceals the cameras, carries an inertial measurement unit, and vents cooling air through the neck. On its own it is a packaging patent for a robot's face. Read against what else the company has recently had issued, it is the last unclaimed region of a machine whose other parts are already covered.

Start below the neck and the granted footprint fills in quickly. The hands are covered by US12605824B2, a humanoid-robot grant describing a wrist and end effector with a multi-degree-of-freedom finger assembly and a thumb with at least three degrees of freedom — dexterity is where manipulation value lives, and the claim reaches for it directly. The torso is covered by US12583126B2, directed to a central region whose actuators give the upper body pitch, roll, and yaw relative to the legs. The legs are covered by US12611766B2, an "advanced kinematics" grant describing a hip-flex-and-knee actuator arrangement, including the design choice to omit a distinct torso-pitch actuator and let the hip actuators carry that motion. Head, hands, torso, hips, knees: the mechanical machine is spoken for.

The software is granted too

What separates this cluster from a pure hardware portfolio is that the control layer is issued alongside the metal. US12638859B2, "Bipedal action model for humanoid robot," is directed to a robot providing at least 30 degrees of freedom with a hierarchical learned model — a Beta model that reads multimodal input into a token sequence and an Alpha model that outputs continuous action chunks of future joint states — plus a low-level controller that turns those chunks into actuator signals. A companion grant, US12578733B2, describes the alpha and beta models deployed across separate GPUs and jointly trained, and US12611767B2 covers an efficient inverse-kinematics control method that solves for target joint parameters within a safety-related limit. The company is having both the body and the policy that drives it issued as patents in the same window.

The present disclosure provides a humanoid robot system comprising a mechanical structure including a torso, two arms, and two legs providing at least 30 degrees of freedom, actuators coupled to the degrees of freedom, a sensor suite comprising at least one camera and proprioceptive sensors including joint encoders and an inertial measurement unit, a computing system comprising at least one processor and memory storing instructions which, when executed, implement a hierarchical bipedal action model including a Beta model configured to receive multimodal input data and generate a token sequence indicative of task intent and environmental state, and an Alpha model configured to condition on the token sequence and current robot pose data to output continuous action chunks comprising sequences of future target joint states over a finite horizon, and a low-level controller configured to convert the continuous action chunks into actuator control signals for execution.— Bipedal action model for humanoid robot, US12638859B2

What the pattern suggests

For a privately held, capital-intensive humanoid company, the composition of the granted set is the readable signal. A firm can file broadly at low cost; grants are the subset that survived examination, and having them arrive across head, hands, torso, legs, and the learned control model at once is consistent with a company treating the entire robot — not one hero component — as the thing to protect. The head grant fits that pattern rather than breaking it: the same three inventors named on US12667974B2 also appear on the earlier head-assembly grants US12403611B2 and US12365094B2, so the head has been prosecuted as a subsystem across multiple issued patents while the rest of the body was covered in parallel. The direction the cluster points is breadth: mechanical enclosure, actuation, dexterity, and autonomy software all being converted from applications into granted rights.

The commercial context is worth stating plainly and without inference beyond the record. Humanoid platforms are a pre-revenue category for most entrants, funded by private capital rather than product cash flow, and the patent record is one of the few public, dated artifacts a private company leaves behind. It shows where research money went roughly a couple of years earlier and what cleared examination. It does not show units shipped, contracts booked, or a valuation — a granted patent is intent and disclosure that survived review, not a fielded product or an order. Reading this cluster as a business signal means reading the shape of the disclosure, not a balance sheet the company has not published.

With those limits fixed, the throughline is legible. The June 30 head grant is the newest entry in a granted body that now runs from the silhouette of the face down through the hands, the torso, the hips and knees, and into the learned model that coordinates all of it. A company having the whole machine issued — enclosure, actuation, and policy together — is telling you, in the one public ledger it cannot help but keep, that its roadmap is the complete humanoid rather than any single part of it. The receipts for that ambition, when and if they arrive as revenue, will have to be read elsewhere; the patent record simply shows the ambition is being built out across the entire robot at once.