Why humanoid robot production IP is a critical bottleneck right now

Annual filings in the humanoid robot manufacturing scale-up space peaked at 75 patent families in 2017, then moderated through the early 2020s before rebounding to 62 in 2023. The most recent years — 2024 and 2025 — show counts of 42 and 57 respectively, but these figures should be read with caution: publication lag means filings from the past 18 to 24 months are systematically under-counted at the time of this analysis. The lifecycle assessment classifies this space as mature, with annual filings plateaued near their historic peak and a recent growth rate of only 2%. That maturity reading is significant: it implies that the foundational claims in bipedal locomotion, manipulator control, and production-line integration have largely been staked, and new entrants face a denser prior-art environment than existed even five years ago.

The technology composition of the corpus reveals a meaningful concentration in robot mechanics and control, but also a striking breadth of adjacent domains. The B25J branch — manipulators and robots — accounts for 179 patents, the largest single cluster. Below it, the next largest branches are B29C (shaping of plastics, 80 patents) and B29D (62 patents), reflecting the materials and forming processes needed to manufacture lightweight robot structures. B65G (materials handling and conveyors, 57 patents) and B65B (packaging machinery, 44 patents) point to the production-line integration challenge: getting robots off an assembly line and into fulfillment environments requires as much IP around the surrounding logistics infrastructure as around the robots themselves. Further down, G06F (digital data processing, 35 patents), H02K (electric motors and generators, 35 patents), G05D (control of non-electric variables, 34 patents), G05B (control and regulating systems, 33 patents), and G06N (AI model computing, 25 patents) together outline the software, actuator, and control stack that underpins real-world deployment.

The core technology sub-battles in humanoid robot scale-up

Beneath the headline production-rate claims, five distinct IP sub-battles are shaping which companies will actually own the manufacturing scale-up stack for humanoid robots.

The IP landscape: who holds what in humanoid robot manufacturing

The top applicant by patent family count is Pirelli Tyre SPA with 78 patent families, a figure that reflects its broader industrial manufacturing IP rather than humanoid-specific robotics — its filings concentrate in B29D (materials forming), G01M (testing), and G01N (material analysis and testing). The second-ranked applicant, Evolved by Nature Inc., holds 31 patent families concentrated in textile chemistry (D06M) and polymer compositions, again reflecting the breadth of the search corpus. Among applicants with clearly humanoid-relevant portfolios, SoftBank Group Corp holds 21 patent families concentrated almost entirely in B25J (manipulators and robots), Figure AI Inc. holds 18 patent families with the most on-topic bipedal filing in the corpus, Apptronik Inc. holds 15 patent families, and AI Inc. holds 12. Tesla Inc. also appears with 18 patent families, with its filings concentrated in G06V (image and video recognition), B60W (hybrid vehicle control), and G06T (image data processing) — reflecting technology transfer from its autonomous-vehicle programs into robotics. The top five filers account for 30% of the hundred largest filers’ combined total, indicating that no single entity has achieved overwhelming dominance and the landscape remains competitive.

Applicant momentum data reveals a split between established incumbents that are slowing and new entrants that are accelerating. SoftBank Group Corp shows a recent filing count of 15 classified as a new entrant in this period, as does Tesla Inc. with 13 recent filings and Evolved by Nature Inc. with 13. By contrast, LK Lab SRL shows a trend of minus 90% in recent filings, and another applicant shows a minus 69% decline — suggesting that some earlier-active filers have wound down their activity. The concentration of new-entrant momentum among SoftBank and Tesla is strategically notable: both are building production-scale humanoid programs and appear to be staking IP positions contemporaneously with their hardware announcements.

Co-filing activity in the corpus is limited but informative. Pirelli Tyre SPA has filed jointly with Politecnico di Torino on 4 occasions, linking a leading industrial manufacturer to a major Italian engineering university. Evolved by Nature Inc. has 2 co-filings with Silk Therapeutics Inc., and Nike Innovate CV has 2 co-filings with Nike Inc. These co-filing relationships are reported as recorded in the patent data; no broader alliance or joint-venture structure should be inferred beyond the filings themselves. The overall low level of co-filing suggests that humanoid robot manufacturing IP is being developed predominantly within individual organizations rather than through formal research partnerships, at least as reflected in this corpus.

Geographically, the United States dominates with 207 patent records, followed by Europe via the EPO with 164 and WIPO PCT applications with 84. Germany accounts for 43 records and Austria for 40, reflecting the strength of European industrial automation and motor-technology applicants. Italy holds 29 records, consistent with Pirelli and other Italian industrial filers. China registers only 13 records in this corpus — a figure that likely understates actual Chinese activity in humanoid robotics and may reflect the specific search scope and jurisdiction coverage of this analysis rather than a genuine absence of Chinese filings in the broader field. South Korea appears with 3 records and Japan with 1, both substantially below what their domestic robotics industries might suggest, again pointing to scope and query-specificity effects.

Foundational and most-cited patents

The most-cited filings in scope anchor the field; citation counts accrue over time, so this list favours older, broadly-cited work. Open any row to see it in Eureka.

Strategic implications for IP and R&D teams

• The core robot-control stack is heavily cited prior art. Navigation, SLAM, and obstacle-recognition patents in this corpus carry hundreds of forward citations each. Any team building production-scale humanoid navigation systems should conduct thorough freedom-to-operate analysis before committing to an architecture that may read on this dense prior-art layer.
• Figure AI’s bipedal action model patent is the most directly on-topic recent filing. US20260064125A1 by Figure AI Inc. discloses a dual-GPU onboard inference architecture for humanoid robots that processes natural language commands — a filing that sits at the intersection of hardware and AI deployment, the exact combination competitors are racing to commercialize. Monitoring continuation filings and citation trees from this patent is a high-priority intelligence task.
• Actuator and motor IP may be under-indexed in humanoid-specific searches. The H02K (electric motors and generators) branch holds only 35 patents in this corpus, modest relative to the centrality of actuator design to bipedal locomotion. IP teams should supplement this analysis with broader industrial motor and actuation searches to avoid missing relevant prior art filed under non-robotics classifications.
• China’s low record count warrants a dedicated search. China appears with only 13 patent records in this corpus, which is almost certainly an artifact of the search scope rather than a true measure of Chinese humanoid robot IP activity. Given the scale of domestic Chinese investment in humanoid robotics programs, a separate CNIPA-focused search is strongly recommended before drawing any competitive conclusions about the Eastern patent landscape.