Count the actuators and the story changes, but look at the foot and it changes again. US11738452B1, granted to Sarcos in August 2023, is "Sole with various compliant regions for robots" — a patent about the least-filmed part of a humanoid, the bottom of its foot.
The B25J filing tells a different story than the keynote. Classified under B62D 57/032 (legged locomotion) and B25J 9/162, the patent describes a sole whose regions have deliberately different compliance — some stiff, some yielding — so the foot manages ground contact passively, the way a human foot's heel, arch and ball do different jobs.
Here is the insight that demos hide: balance is partly a mechanical problem solved below the ankle, before any controller intervenes. A foot with engineered compliance absorbs the shock of landing and conforms to uneven ground passively, taking load off the control system. Stiffness everywhere makes a rigid, clattering robot; compliance everywhere makes a mushy, imprecise one. The varied regions are the compromise.
This pairs with the ankle-region work other humanoid players have patented. Together they make a consistent argument: the load path from torso to floor — ankle, then sole, then ground — is where standing is actually won, and it is mechanical as much as algorithmic. Dexterity is a battery problem before it's an AI problem, and balance is a foot problem before it's a control problem.
The honest limit, as always, is that a patent describes structure, not durability. A compliant sole that works in a lab has to survive thousands of footfalls a shift on real factory floors, and the patent does not disclose that wear life. The mechanism is sound; the duty cycle is the open question.
For readers auditing humanoid claims, Sarcos' sole patent is the reminder to watch the feet. When a company files on the bottom of its robot's foot, it is telling you — more honestly than any upper-body manipulation video — that it knows where the unsolved balance problem lives.