Assistive Robotics for People with Special Needs — Patent Landscape

Executive Summary

What the Cooperative Patent Classification calls assistive robotics, counted at family level, 2000–2023.

The Cooperative Patent Classification contains ten places whose own text names the purpose directly: robots that aid disabled people, rehabilitation and therapy robots, exoskeletons, powered walking aids, nursing manipulators. Taken together they hold 9,450 DOCDB patent families, filed as 17,631 applications at 49 patent authorities, with an earliest filing year between 2000 and 2023. This report is the anatomy of that set.

Activity is heavily back-loaded. Families with a first filing in 2000–2011 number 917; the 2012–2017 period holds 2,474 and 2018–2023 holds 6,059 — almost two thirds of the whole set in the last six years of the window. Annual filings run from 30 families in 2000 to a peak of 1,080 in 2021, with 944 in 2023. Between 2012 and 2022 the series compounds at 17.8 % a year.

The four concepts are not equally weighted. Exoskeletons are the largest at 4,254 families, followed by rehabilitation and therapy robots (3,065) and robotic walking and gait aids (2,367); personal assistance and nursing robots remain a comparatively small 956. Because a family can carry codes from more than one concept, the four figures sum to 10,642 against 9,450 distinct families, an overlap of 1,192 concept memberships.

Geographically the field has recentred. In 2000–2011, 442 of 917 families included a US filing and 97 a Chinese one. In 2018–2023, 4,096 of 6,059 families include a CN filing against 1,266 with a US filing. Over the whole window China leads on 5,444 families, ahead of the United States (2,555) and Korea (2,219). The share of families filed at two or more authorities moves the other way: 40.2 % in 2000–2011, 33.8 % in 2012–2017, 21.0 % in 2018–2023.

On the applicant side, no single organisation dominates. After consolidating name variants, the largest portfolio is Samsung Electronics with 242 families, ahead of the Hyundai Motor Group (139), Toyota Motor (119) and Honda Motor (112). Public research is unusually prominent for a device field: 2,815 families carry a university applicant and 576 a public research applicant, against 3,979 with a company applicant.

The set is defined by classification, not by keyword, and that boundary is visible in the data. A title-keyword search over the same window returns 10,970 further families that name exoskeletons, rehabilitation robots, walking assistance or gait training but carry none of the ten defining codes. A sample of those families contains both genuinely robotic assistive devices classified elsewhere and passive walking aids that this scope excludes by design. The 9,450 figure is therefore a precise, reproducible cut of the field — not a census of everything that could reasonably be called assistive robotics.

Filing Trend 2000–2023

DOCDB patent families by earliest filing year, and the share of them filed at two or more authorities.

Annual filings grow from 30 families in 2000 to a peak of 1,080 in 2021. Across the interior window 2012–2022 the series compounds at 17.8 % per year (200 to 1,029 families). Two thirds of the whole set — 6,059 of 9,450 families — has an earliest filing year of 2018 or later.

The step between 2015 and 2016 is a discontinuity rather than a trend: 271 families to 722, an increase of 166 % in one year, larger than any adjacent change. It falls in the period during which China became the largest filing cohort in this field, but this data does not establish a cause and the jump should be read as a break in the series.

Internationalisation moves against volume. The share of families filed at two or more authorities is 40.2 % for first filings in 2000–2011 (369 of 917), 33.8 % for 2012–2017 (835 of 2,474) and 21.0 % for 2018–2023 (1,275 of 6,059). Across the whole window 2,479 families (26.2 %) reach two or more authorities and 1,790 (18.9 %) include a WO application.

Each family is counted once, in its earliest filing year, so the annual column sums exactly to 9,450. Filings up to 2023 are substantially complete at this edition — publication of a 2023 filing is normally finished well before a Spring 2026 snapshot — but the last years of any window still fill up slightly, so no growth statement here rests on 2023 alone.

Data table: Filing trend 2000–2023

Year	Patent families	Filed at 2+ authorities	International share
2000	30	11	36.7%
2001	34	10	29.4%
2002	39	20	51.3%
2003	45	19	42.2%
2004	52	21	40.4%
2005	61	25	41.0%
2006	73	30	41.1%
2007	88	35	39.8%
2008	114	58	50.9%
2009	108	52	48.1%
2010	123	44	35.8%
2011	150	44	29.3%
2012	200	62	31.0%
2013	215	101	47.0%
2014	206	115	55.8%
2015	271	150	55.4%
2016	722	197	27.3%
2017	860	210	24.4%
2018	927	224	24.2%
2019	1,004	274	27.3%
2020	1,075	217	20.2%
2021	1,080	207	19.2%
2022	1,029	198	19.2%
2023	944	155	16.4%
Total	9,450	2,479	26.2%

Technology Profile

Four robotic concepts, the individual CPC places behind them, and the technical fields the families also sit in.

The four concepts

The ten defining CPC codes group into four concepts. The grouping is interpretive — it follows what the code titles describe — and the concepts overlap: a powered gait trainer can be both an exoskeleton and a rehabilitation robot.

Exoskeletons are the largest concept at 4,254 families, 45.0 % of the set, and the one that grew hardest: 255 families in 2000–2011 against 2,911 in 2018–2023. Robotic walking and gait aids show the opposite profile — the largest concept in the early period at 574 families, and the slowest multiple since, reaching 1,115 in 2018–2023. Personal assistance and nursing robots remain the smallest strand throughout (956 families in total).

Concepts are not mutually exclusive. The four concept totals sum to 10,642 against 9,450 distinct families, an overlap of 1,192 concept memberships: families that carry defining codes from more than one concept are counted in each. Concept figures must not be added together.

Data table: Patent families per concept and period

Concept	2000–2011	2012–2017	2018–2023	Total families	Share of 9,450
Exoskeletons	255	1,088	2,911	4,254	45.0%
Rehabilitation and therapy robots	186	765	2,114	3,065	32.4%
Robotic walking and gait aids	574	678	1,115	2,367	25.0%
Personal assistance and nursing robots	46	315	595	956	10.1%
Distinct families	917	2,474	6,059	9,450	100%

Families per defining code

Below the concept level, the set is carried by six CPC places. The four remaining defining codes are robotics tags in the G05B2219 indexing scheme and each holds fewer than 250 families, below the reporting threshold used here.

Data table: Families per defining CPC code (codes with 250 families or more)

CPC code	Title	Concept	Families
`B25J9/0006`	Exoskeletons, i.e. resembling a human figure	Exoskeletons	4,225
`A61H1/0262`	Walking movement; appliances for aiding disabled persons to walk	Rehabilitation and therapy robots	1,767
`A61H3/008`	Walking aids using suspension devices supporting the body upright	Robotic walking and gait aids	1,496
`A61H2201/1659`	Free spatial automatic movement of the therapy interface within a working area	Rehabilitation and therapy robots	1,485
`A61H2003/043`	Wheeled walking aid with a drive mechanism	Robotic walking and gait aids	973
`B25J11/009`	Manipulators for nursing, e.g. carrying sick persons, pushing wheelchairs	Personal assistance and nursing robots	775

What else the families are classified as

A family carries many CPC symbols. Reading the ones outside the defining set shows which technical fields assistive robotics actually borrows from: 8,796 of 9,450 families (93.1 %) carry at least one CPC code beyond the ten that define the scope.

The set sits astride two subclasses: A61H (physical therapy apparatus) on 5,925 families and B25J (manipulators) on 5,377. That split is the field itself — a medical-device tradition and a robotics tradition converging on the same devices. Beyond them the profile is device-oriented rather than software-oriented: physical training apparatus (A63B, 1,037), diagnosis and surgery (A61B, 832), prostheses and orthopaedic devices (A61F, 762) and patient conveyances (A61G, 717) all outweigh healthcare informatics (G16H, 304) and digital data processing (G06F, 297).

253 families carry a Y02A tag — the CPC-only tagging section for adaptation to climate change, which includes health-adaptation technologies. Y tags have no IPC equivalent, so counts based on them under-represent offices that do not assign CPC.

Subclass totals exceed the family total. One family carries several CPC symbols, so the ten subclass figures above sum to well over 9,450. They are counts of families per subclass, not parts of a whole, and are shown as bars rather than as a pie for that reason.

Data table: Co-occurring CPC subclasses

Subclass	Description	Families	of which 2018–2023	Share of 9,450
`A61H`	Physical therapy apparatus	5,925	3,559	62.7%
`B25J`	Manipulators	5,377	3,615	56.9%
`A63B`	Apparatus for physical training, gymnastics	1,037	533	11.0%
`A61B`	Diagnosis; surgery; identification	832	504	8.8%
`A61F`	Prostheses; orthopaedic devices	762	368	8.1%
`A61G`	Transport and personal conveyances for patients	717	445	7.6%
`G05B`	Control or regulating systems in general	420	240	4.4%
`G16H`	Healthcare informatics	304	231	3.2%
`G06F`	Electric digital data processing	297	223	3.1%
`Y02A`	Technologies for adaptation to climate change	253	233	2.7%

Geographic Analysis

Where the families are filed, and how that footprint moved across three periods.

Top filing offices

Shift by period

Over the whole window 5,444 families include a Chinese filing — 57.6 % of the set — ahead of the United States (2,555, 27.0 %) and Korea (2,219, 23.5 %). The ordering is recent. In 2000–2011 the United States appeared in 442 of 917 families (48.2 %) and China in 97 (10.6 %); in 2018–2023 China appears in 4,096 of 6,059 families (67.6 %) and the United States in 1,266 (20.9 %).

Japan is the only major office whose absolute count falls between the last two periods, from 390 families in 2012–2017 to 375 in 2018–2023. Every other office in the chart grows in absolute terms; what changes is their share of a much larger total.

Office totals exceed the family total. A family filed in China, the United States and at the EPO is counted at each of the three offices, so the columns sum above 9,450 (and above the 917 / 2,474 / 6,059 period totals). WO counts families with a PCT application, which is a route rather than a market.

Why applicant origin is not shown

The complementary question — where the applicants are based, rather than where they file — is not answerable from this scope with acceptable confidence. 5,792 of the 9,343 families with an applicant record (62 %) carry a blank country code on every applicant, and Chinese applicants are coded for only 275 families against the 5,444 families with a Chinese filing. Any origin chart built on the coded remainder would describe the coding, not the field. Filing authority is used throughout this report instead, and it answers a different question: which markets the field protects in.

Data table: Top filing offices — patent families 2000–2023 (backs the "Top filing offices" chart)

Office	2000–2011	2012–2017	2018–2023	Total families	Share of 9,450
CN — China	97	1,251	4,096	5,444	57.6%
US — United States	442	847	1,266	2,555	27.0%
KR — Korea	342	735	1,142	2,219	23.5%
WO — PCT	298	570	922	1,790	18.9%
EP — European Patent Office	194	459	559	1,212	12.8%
JP — Japan	204	390	375	969	10.3%
DE — Germany	117	144	184	445	4.7%
CA — Canada	76	120	124	320	3.4%
AU — Australia	57	80	80	217	2.3%
FR — France	29	55	74	158	1.7%
ES — Spain	35	69	49	153	1.6%
IT — Italy	8	40	51	99	1.0%
Distinct families	917	2,474	6,059	9,450	100%

Data table: Shift by period — patent families per office (backs the "Shift by period" chart)

Office	2000–2011	2012–2017	2018–2023
CN — China	97	1,251	4,096
US — United States	442	847	1,266
KR — Korea	342	735	1,142
WO — PCT	298	570	922
EP — European Patent Office	194	459	559
JP — Japan	204	390	375
DE — Germany	117	144	184

The seven offices charted here are those with the largest presence in the set. A family filed at several of them is counted at each, so the columns sum above the 917 / 2,474 / 6,059 distinct-family totals for the three periods.

Top Applicants

Portfolios after consolidating harmonised name variants, and the sector composition of the field.

The field has no owner. Samsung Electronics holds the largest portfolio at 242 families — 2.6 % of the set — and the fifteen largest portfolios together account for at most 1,307 of 9,450 families, under 14 %, before any deduplication of jointly filed families. Assistive robotics is patented by many small holders rather than by a handful of large ones.

The composition of the leaders changes across the window. Honda Motor is the only top portfolio concentrated in the earliest period, with 86 of its 112 families first filed before 2012. Samsung (156 of 242), LG Electronics (75 of 90), the Shenzhen Institutes of Advanced Technology (61 of 66) and four Chinese universities have most or all of their families in 2018–2023.

Name harmonisation is imperfect, and the counts are lower bounds. PATSTAT's harmonised applicant name splits several of these organisations across variants — Toyota appears under 18, Panasonic under 19, the Hyundai group under 15 and Samsung under 11. The table below groups the variants that resolve to the same organisation — including variants written in Japanese and Chinese script — and counts each family once per group, so the figures are distinct-family counts and not sums of the variant rows. Variants that no rule catches remain separate, which makes every count here a lower bound. The country column reports only what PATSTAT codes; it is blank for many applicants and is not inferred from the organisation's name.

Data table: Fifteen largest consolidated portfolios (backs the chart above)

#	Applicant group	Sector	Country (PATSTAT-coded)	Name variants	2000–2011	2012–2017	2018–2023	Families
1	Samsung Electronics	Company	KR	11	11	75	156	242
2	Hyundai Motor Group	Company	KR	15	10	63	66	139
3	Toyota Motor	Company	JP	18	8	37	74	119
4	Honda Motor	Company	JP	8	86	11	15	112
5	LG Electronics	Company	KR	3	6	9	75	90
6	National Rehabilitation Center	Public research	KR	1	3	33	34	70
7	KAIST	University	KR	1	8	22	38	68
8	Shenzhen Institutes of Advanced Technology, CAS	Public research	CN	4	0	5	61	66
9	Hebei University of Technology	University	—	1	0	13	53	66
10	Panasonic	Company	JP	19	14	29	17	60
11	Hefei University of Technology	University	—	1	0	13	47	60
12	Harbin Institute of Technology	University	—	1	0	21	37	58
13	Agency for Defense Development	Public research	KR	1	0	27	29	56
14	Hangzhou Chengtian Technology Development	Company	—	1	0	4	48	52
15	JTEKT	Company	JP	4	0	8	41	49

Sector composition

9,343 of the 9,450 families carry at least one applicant record. Sectors are taken from PATSTAT's psn_sector and are not exclusive: a family co-filed by a company and a university counts in both.

Universities appear on 2,815 families, public research bodies on 576 and organisations classified as both on a further 147, against 3,979 families with a company applicant. For a field of physical devices that is an unusually academic profile, and it is concentrated in the recent period — 1,938 of the 2,815 university families were first filed in 2018–2023. 1,595 families carry an individual inventor-applicant, a share that points to a low barrier to entry for mechanical assistive devices.

1,319 families (14.1 % of those with an applicant record) include an application with more than one applicant.

Data table: Applicant sectors

PATSTAT sector	Families	Share of 9,343
COMPANY	3,979	42.6%
UNIVERSITY	2,815	30.1%
INDIVIDUAL	1,595	17.1%
UNKNOWN	1,399	15.0%
GOV NON-PROFIT	576	6.2%
GOV NON-PROFIT UNIVERSITY	147	1.6%
HOSPITAL	133	1.4%
(not stated)	74	0.8%

Sectors are non-exclusive; the column sums to 10,718 against 9,343 families with an applicant record.

Grant Rates

Outcomes for the cohort old enough to be mostly decided: families with an earliest filing year up to 2018.

For applications belonging to families first filed between 2000 and 2018, 69.3 % were granted (5,927 of 8,547, PCT applications excluded). The spread across offices is wide: Korea at 79.8 % and the United States at 78.1 % against China at 60.6 % and Germany at 44.8 %. At family level the same cohort shows 3,097 of 4,318 families (71.7 %) with at least one granted member.

China combines the largest volume in this field with the lowest grant rate among the major offices — 1,275 grants from 2,103 applications in the decided cohort. Because Chinese filings dominate the recent years, that difference also pulls the overall rate down over time.

Pendency caveat. granted = 'Y' means granted as of this snapshot. Families first filed in recent years have not had time to be examined, so their apparent grant share measures examination speed as much as success: 83.4 % of families first filed in 2015 have a granted member, against 48.3 % for 2022 and 29.4 % for 2023. Every rate quoted above is restricted to the 2000–2018 cohort for that reason, and no rate should be read from the last filing years. The series is not perfectly monotonic — 2019 sits above 2020 and 2021 — so pendency is not the only influence on it, and this data does not isolate the others.

PCT applications are excluded from the headline rate: WO applications are not granted, and including their 1,034 applications with 32 grants would depress the figure without meaning.

Data table: Grant outcomes by office, families first filed 2000–2018

Office	Applications	Granted	Grant rate
KR — Korea	1,308	1,044	79.8%
US — United States	1,960	1,530	78.1%
JP — Japan	747	561	75.1%
EP — European Patent Office	859	586	68.2%
CN — China	2,103	1,275	60.6%
AU — Australia	160	94	58.8%
CA — Canada	222	121	54.5%
DE — Germany	315	141	44.8%
ES — Spain	123	123	100.0%
WO — PCT (excluded)	1,034	32	3.1%
All offices excl. WO	8,547	5,927	69.3%

Spain's 100 % reflects a national procedure in which the counted applications all issued; with 123 applications it is not comparable to the large offices.

Data table: Share of families with a granted member, by earliest filing year

Earliest filing year	Families	With a granted member	Share
2010	123	102	82.9%
2015	271	226	83.4%
2019	1,004	674	67.1%
2020	1,075	676	62.9%
2021	1,080	630	58.3%
2022	1,029	497	48.3%
2023	944	278	29.4%

Methodology

Search strategy, data basis, and what this report does and does not cover.

Updated 15 August 2026 against EPO PATSTAT Global, Spring 2026 (patstat-mtc.patstat_2026a), with IPC edition 20260101 and CPC edition 202608. The classification scope was completed by subtree expansion, by union with the predecessor symbols from the reclassification chain, and by retaining deprecated symbols; on this code set all three were applied and verified to add no further symbols.

Search strategy, counting rules and data source

Search strategy

The scope was built from the classification scheme itself, using three independent entry points into the CPC and IPC, and then narrowed by a precision test.

Catchword index. WIPO's natural-language index into the IPC was searched for robot, manipulator, exoskeleton, walking aid, invalid, disabled and handicapped. It anchors the subject at A61H3/00 (walking aids for invalids), B25J (robots in general), A61G3/00 and A61G5/00 (vehicles and chairs for invalids).
Title breadcrumb. title_full on the classification views carries the ancestor chain, so subgroup titles that are fragments in isolation stay findable. Singular and plural forms were searched separately and unioned. This is where the codes that name assistance explicitly surface, including four robotics tags in the G05B2219 indexing scheme.
Definition texts. The CPC scope prose was searched for assistive, disabled persons, handicapped, gait training, walking assistance and rehabilitation robot. It confirmed A61H3/00 and A61H3/008 and added A61H1/0237.
Completion. Every seed was expanded to its subtree by recursion on the parent relation — never by symbol prefix, which pulls in siblings rather than descendants — unioned with its predecessors from the reclassification chain, and left unfiltered on is_active so that retired symbols keep their documents. All ten symbols turned out to be leaves with no predecessor and none is deprecated, so the completed set equals the seed set.

The ten defining CPC codes

CPC code	Title	Concept
`B25J9/0006`	Exoskeletons, i.e. resembling a human figure	Exoskeletons
`G05B2219/40305`	Exoskeleton, human robot interaction, extenders	Exoskeletons
`A61H1/0262`	Walking movement; appliances for aiding disabled persons to walk	Rehabilitation and therapy robots
`A61H2201/1659`	Free spatial automatic movement of the therapy interface within a working area	Rehabilitation and therapy robots
`G05B2219/45109`	Exercise, coordination, therapy, rehabilitation robot for disabled	Rehabilitation and therapy robots
`A61H3/008`	Walking aids using suspension devices supporting the body upright	Robotic walking and gait aids
`A61H2003/043`	Wheeled walking aid with a drive mechanism	Robotic walking and gait aids
`B25J11/009`	Manipulators for nursing, e.g. carrying sick persons, pushing wheelchairs	Personal assistance and nursing robots
`G05B2219/45108`	Aid, robot for aid to, assist human disabled	Personal assistance and nursing robots
`G05B2219/40411`	Robot assists human in non-industrial environment like home	Personal assistance and nursing robots

Why CPC only

None of the ten symbols exists in the IPC. The scheme bridge maps four of them only to far broader parents — A61H1/02, A61H3/00, B25J9/00, B25J11/00 — and six to nothing at all.
For this subject CPC is therefore not a preference over IPC but the only scheme that resolves the technology at all. The consequence is a coverage limit: offices that do not assign CPC are under-represented, and every count here rests on tls224_appln_cpc.

How the scope was narrowed

A wider candidate scope was tested first: the four concepts plus generic collaborative robotics, legged locomotion, embodied artificial life and home and service robots. It held 42,616 families.
A stratified sample of titles from that candidate scope was read by hand. The humanoid and artificial-life strand returned neural-network hardware, power-distribution control and image-analysis patents; the service strand returned cleaning and rescue robots. Twenty of twenty-four randomly drawn families were off-subject.
The candidate scope was rejected and the code set reduced to the ten places whose own text names assistance, rehabilitation or nursing. A fresh sample of twenty-five families from the final scope contains four off-subject titles — a residual noise rate of 16 %, inside the one-in-five limit that governs whether a scope may be used.

Family counting

The unit is the DOCDB patent family. An invention filed in China, the United States and at the EPO is one family, not three.
A family enters the scope if any of its applications carries one of the ten codes. Its year is the earliest filing year across all of its applications, so a family is counted in one year only and the annual column sums exactly to 9,450.
Once a family is in scope, all of its applications are used for the office, grant and applicant analysis — not only the applications that carry a defining code.
Pseudo-families (docdb_family_id = 0) and design applications (appln_kind = 'D') are excluded.
Applicants are taken with applt_seq_nr > 0; inventors are not analysed in this report.

Multi-assignment

Concepts overlap: the four concept totals sum to 10,642 against 9,450 distinct families.
Offices overlap: a family filed at several offices is counted at each of them.
CPC subclasses overlap: a family carries many symbols, so the subclass column sums far above the family total.
Applicant sectors overlap: the sector column sums to 10,718 against 9,343 families with an applicant record.
Sub-totals in this report therefore exceed the headline figure by design, and are never drawn as parts of a whole.

Data source and toolchain

EPO PATSTAT Global, Spring 2026 edition (patstat-mtc.patstat_2026a) on Google BigQuery.
Classification views patstat-mtc.classification.dpt_cpc, dpt_ipc, dpt_catchword, dpt_scheme_map and the reclassification chains, at IPC edition 20260101 and CPC edition 202608.
Queried through the patstat-mcp server; analysis and figures produced with Claude AI. Every number in this report is traceable to a numbered statement in queries.sql.

Scope Limitations

The scope is precision-first, and a keyword layer finds a comparable volume outside it. A title search over the same window for exoskeleton, rehabilitation robot, walking assist, gait training, nursing robot, assistive robot and walking aid returns 14,219 families, of which 3,249 are in the scope and 10,970 are not. A sample of twenty-five of those 10,970 contains two distinct kinds of family: genuinely robotic assistive devices classified under codes outside the defining set — hand and wrist exoskeleton rehabilitation trainers, gait-training robot leg supports, soft rehabilitation robot gloves — and passive walking aids that this scope excludes by design. Widening the code set to reach them was tested and rejected because it destroyed precision. The 9,450 figure is a reproducible cut of the field, not a census of it.
Non-robotic assistive devices are outside the scope by design. Plain orthopaedic braces (A61F5/01), passive exercise apparatus (A61H1/02) and unpowered wheeled walking aids (A61H3/04) are not counted. Measured together, those three code groups hold 104,555 families in this window — an order of magnitude larger than this scope, and a different subject.
Powered wheelchairs and patient-transport vehicles are not counted. A61G is a co-occurring subclass on 717 families but is not part of the defining set.
CPC-only coverage. Offices that do not assign CPC are under-represented; no IPC equivalent of the ten codes exists to compensate.
Applicant origin is not reported. 62 % of families with an applicant record carry a blank country code on every applicant, and Chinese applicants are coded for only 275 families against 5,444 families with a Chinese filing.
Applicant counts are lower bounds. Harmonised names split organisations across variants. The consolidation in this report merges the variants that resolve to the same organisation, in Latin, Japanese and Chinese script, but any variant no rule catches stays separate.
Grant rates are restricted to the 2000–2018 cohort and exclude PCT applications. Later filings are still under examination.
No external benchmark was consulted. The family count has not been checked against a third-party database, so it carries only internal consistency, not external corroboration.

Glossary — patent terms explained

Patent family (DOCDB): A group of patent applications protecting the same invention in different countries. Counted once, so that an internationally filed invention does not inflate the totals.
Earliest filing year: The lowest filing year across all applications of a family. Used as the family's single position on the time axis.
International family: A family with applications at two or more patent authorities — a signal that the applicant expected a market beyond the home one.
CPC (Cooperative Patent Classification): The classification scheme jointly maintained by the EPO and the USPTO. Finer than the IPC, and the only scheme containing the ten places this report is built on.
Indexing code (the 2000-series): A CPC symbol such as A61H2201/1659 or G05B2219/45108 that tags an additional aspect of a document rather than placing it in the scheme. Assistive robotics is largely described by such tags.
Subtree expansion: Following a classification symbol down to all of its descendants. Necessary because subject matter belonging to a subgroup is never classified into its parent.
Predecessor union: Adding the symbols a place was carved out of, so that a time series measures invention rather than the progress of reclassification.
Grant rate: Granted applications divided by all applications in a cohort. Meaningful only for filings old enough to have been examined.
PCT / WO application: An international application filed under the Patent Cooperation Treaty. It reserves the option of protection in many countries; it is not itself granted.

Report files

The complete SQL behind this report ships with it as queries.sql, in execution order and annotated with the result each statement produced. Every figure in the text, in the tables and in the chart configurations traces back to one of those statements. The file names the PATSTAT edition, the IPC and CPC editions and the classification dataset, so the analysis can be re-run and compared like for like.

Like what you see?

This report was built with a fully reproducible pipeline: EPO PATSTAT Global on BigQuery, a custom MCP server, and Claude AI for analysis and visualization.

Custom Patent ReportNeed a similar analysis for your technology area, company, or research question? We build tailored patent intelligence reports.

The ToolchainWant to run your own AI-powered patent analyses? We help you set up PATSTAT on BigQuery with MCP server integration.

Arne Krueger — LinkedIn Book a Call depa.tech mtc.berlin