This landscape identifies 31 patent families published between 2015 and mid-2026 that combine humanoid, bipedal or legged robot claims with tactile, force/torque or skin sensor claims under IPC classes B25J13/08, G01L5/00 and B25J19. That is a narrow, specific slice of robotics patenting, not a count of all humanoid robot patents. Filing peaked at 11 families in 2025, and the 2026 figure is still incomplete because publication typically trails filing by about 18 months.

No single company dominates this corpus outright; activity is spread across corporate, research-institute and university assignees. The most visible structured collaboration links a Japanese automaker, its affiliated research institute, and a partner university, who appear together across the strongest co-assignee pairs in the dataset. Recent-year momentum has been mixed to declining across most active filers, so leadership in this niche is still unsettled rather than established.

US20260091509A1, assigned to Figure AI Inc., claims a humanoid robot end effector built around a tactile sensor assembly with a bridge structure of three flexure arms, each combining a sloped and a load section at non-zero angles, with a strain gauge on each sloped section feeding a shared electronics assembly. It is a specific mechanical and sensing architecture, not a claim over tactile sensing generally, so designs using different geometries, different numbers of flexure arms, or non-strain-gauge sensing modalities sit outside its literal scope. Anyone designing a similar three-arm strain-gauge bridge should review its claims closely before finalising geometry.

The clearest gaps relative to core strain-gauge and force/torque claims are in distributed, skin-wide tactile arrays, sensor fusion that ties tactile feedback directly to whole-body balance control, and force sensing tailored to wheel-legged hybrid platforms, which appear in a third of the corpus via B62D classification but with little dedicated sensing claim depth. These are areas with real technical activity in the surrounding literature but thin patent claim density in this specific dataset. That combination — active development, light claiming — is what makes a branch worth checking before assuming it is occupied.

Filings rose from zero in 2017 to a peak of 11 families in 2025, but several previously active assignees show zero filings in the latest year, and the one assignee still filing recorded a steep year-on-year decline. That pattern looks more like an early cluster of entrants followed by a shakeout than a field building steady momentum. The 2026 count is too early in its publication cycle to confirm whether this is a genuine slowdown or simply a reporting lag.