Skip to main content

Autonomous Commercial Scrubber-Dryers

Patent landscape of autonomous floor-scrubbing machines — 2,766 DOCDB families whose earliest filing falls in 2019–2024, defined by the CPC floor-scrubbing subtree (A47L11/26, /28, /29 and their descendants) intersected with the robotic-cleaning indexing scheme A47L2201.

Data: EPO PATSTAT Global, Spring 2026 (patstat_2026a) Classification: IPC 20260101 · CPC 202608 Period: 2019–2024 Author: mtc.berlin
2,766
Patent Families
First filing 2019–2024
5,153
Applications
Across 26 offices
92.7%
Reach the CN Office
2,564 families
57.2%
Grant Rate
2019–2021 cohorts
95
Professional-Segment Families
14 manufacturers

Executive Summary

Who is patenting autonomous floor-scrubbing machines, how the activity developed 2019–2024, and where the professional cleaning-equipment industry sits in that picture.

A scrubber-dryer wets the floor, scrubs it and takes the dirty liquid straight back up. The classification system says exactly that in one place: the CPC floor-scrubbing groups A47L11/26, A47L11/28 and A47L11/29 with their descendants. Crossing those thirteen codes with the robotic-cleaning indexing scheme A47L2201 — the CPC tag for machines that control their own travel — gives the set this report analyses: 2,766 DOCDB patent families whose earliest filing falls between 2019 and 2024, carried by 5,153 individual applications at 26 patent offices.

Filing activity multiplied in the first half of the window. Families with a first filing rose from 177 in 2019 to 620 in 2022, a compound annual growth rate of 51.9 % over those three years. The two trailing years, 561 and 425, sit inside the ~18-month publication delay and are provisional; no growth statement in this report rests on them.

The geography is lopsided in a way that shapes every other reading. 2,564 of the 2,766 families — 92.7 % — have a member at the Chinese office, against 503 with a WO member, 401 with a US member and 333 with an EP member. The PCT route is used by 18.2 % of families. Between the 2019–2021 and 2022–2024 halves of the window the US share of families fell from 21.8 % to 9.2 % and the WO share from 22.5 % to 15.1 %, while the German office moved the other way, 3.4 % to 4.9 %.

The applicants with the most families in scope are broad-market floor-care and appliance brands: Beijing Roborock with 178 families, Shenzhen Silver Star with 169, Dreame Innovation with 141, Gree Electric with 102. Classification cannot separate a machine sold into an office building from one sold into a living room, so this report adds an explicit, manually assigned layer: 14 manufacturers whose business is professional cleaning equipment or commercial cleaning robots account for 95 families, 3.4 % of the scope. Within that group Gaussian Automation holds 28 families and Alfred Kärcher 21; the remaining twelve hold eight or fewer each. That cohort's annual output stayed inside a narrow band all six years — 11, 17, 21, 19, 15, 12 families.

Two boundaries of the search are worth carrying into any reading of the numbers. The autonomy marker A47L2201 exists only in CPC, and 10,669 of the 18,696 families carrying an IPC floor-scrubbing code in this window hold no CPC assignment at all — they cannot be reached by this profile. And a title search for scrubbing plus autonomy terms returns 562 families outside the classification scope. Both figures are stated with their queries in the Methodology section.

The data shows a field that grew roughly threefold between 2019 and 2022 and is filed overwhelmingly in China. Professional floor-care manufacturers hold 95 of 2,766 families — the autonomous scrubbing patent record is written largely by broad-market robot-cleaner brands, and the professional segment appears in it as a small, steady cohort rather than as a growth curve.

Search Profile

How the 2,766 families were selected, and what each classification layer contributes.

The research question is narrow: which patent families describe floor-scrubbing machines that control their own travel, filed for the first time between 2019 and 2024? Three entry points into the classification scheme were used, and they agree on where the subject lives.

  • Catchword index (IPC). WIPO's natural-language index maps „SCRUBBING of floors“ to A47L11/26, A47L11/28 and A47L13/10. The third is a hand implement, not a machine, and stays outside. The same query returns „SCRUBBING of gases“ → B01D47/00, the homonym that a plain keyword search would drag in.
  • Title breadcrumb (CPC). SEARCH(title_full, …) on scrubbing, scrubbers, floor cleaning, cleaning floors, cleaning robot and cleaning robots returns the full A47L11 and A47L13 breadcrumb plus A47L2201/028. Singular and plural forms were searched separately and unioned.
  • Definition texts (CPC). SEARCH(definition_scope, …) on floor scrubbing, scrubber dryer and floor cleaning machine(s) returns A47L11/00 and A47L23/00 only. For this subject the definition layer confirms the title layer and adds no symbol to it.

The code set

Three seed groups were expanded to their full subtree with a recursive walk on parent, which turns 3 codes into 13:

CPC Title Families in scope
A47L11/26Floor-scrubbing machines, hand-driven10
A47L11/28Floor-scrubbing machines, motor-driven914
A47L11/282… having rotary tools748
A47L11/283… the tools being disc brushes267
A47L11/284… having reciprocating tools145
A47L11/29Floor-scrubbing machines characterised by means for taking up dirty liquid118
A47L11/292… having rotary tools201
A47L11/293… the tools being disc brushes47
A47L11/294… having reciprocating tools12
A47L11/30… by suction147
A47L11/302… having rotary tools344
A47L11/305… the tools being disc brushes71
A47L11/307… having reciprocating tools18

The autonomy marker is the indexing scheme A47L2201 — „Robotic cleaning machines, i.e. with automatic control of the travelling movement“ — whose recursive subtree contains 8 symbols (/00, /02, /022, /024, /026, /028, /04, /06). A family enters the scope when it carries at least one code from each of the two sets.

Data table: Classification layer coverage
Layer DOCDB families Relative to the A47L11 layer
A47L11 (any group) — machines for cleaning floors16,121100.0%
A47L2201 (any group) — robotic cleaning machines7,56046.9%
Floor-scrubbing subtree (13 codes)6,71341.6%
A47L11 × A47L22016,60341.0%
Report scope: scrubbing subtree × A47L22012,76617.2%

All figures count DOCDB families whose earliest filing year across all members falls in 2019–2024. Layers overlap; the rows are not additive.

Restricting from „any floor-cleaning machine“ to the floor-scrubbing branch cuts the autonomous set from 6,603 to 2,766 families. A random sample of 22 families read by English title contained no off-topic case — every one described an autonomous or automatically controlled floor-cleaning machine, well inside the 1-in-5 threshold the landscape workflow sets for a usable profile.

Filing Trend 2019–2024

Patent families by their earliest filing year, with the applications those families comprise.

* 2023–2024: incomplete data due to the ~18-month publication delay (PATSTAT Global, Spring 2026, patstat_2026a).

Data table: Filing trend 2019–2024
Year Families Year on year Applications Applications per family
20191776153.47
2020370+109.0%8122.19
2021613+65.7%1,0471.71
2022620+1.1%1,0751.73
2023*561−9.5%9511.70
2024*425−24.2%6531.54
Total2,7665,1531.86

Each family is counted once, in the earliest filing year across all its members, so the yearly figures sum exactly to the total. Applications include every non-design member of a scope family, also those filed outside the 2019–2024 window.

Families with a first filing rose from 177 in 2019 to 620 in 2022 — a compound annual growth rate of 51.9 % over those three years, and a 2021 figure (613) already within 1 percentage point of the 2022 peak. The 2023 and 2024 figures sit inside the publication delay and are provisional; the decline they show is not read here as a change in filing behaviour.

Data quality note. Applications per family fall from 3.47 in 2019 to 1.54 in 2024. This is consistent with the same publication delay: later family members — national-phase entries, continuations, second filings — are published after the priority filing, so recent cohorts are observed earlier in their life. The ratio is not used in this report as a measure of internationalisation.

Technology Profile

Where inside the scrubbing branch and inside the robotic indexing scheme the 2,766 families sit.

Data table: Floor-scrubbing subgroups
CPC Title Families
A47L11/28Floor-scrubbing machines, motor-driven914
A47L11/282… having rotary tools748
A47L11/302… dirty liquid taken up by suction, having rotary tools344
A47L11/283… the tools being disc brushes267
A47L11/292… dirty liquid taken up, having rotary tools201
A47L11/30… dirty liquid taken up by suction147
A47L11/284… having reciprocating tools145
A47L11/29Floor-scrubbing machines characterised by means for taking up dirty liquid118
A47L11/305… by suction, the tools being disc brushes71
A47L11/293… the tools being disc brushes47
A47L11/307… by suction, having reciprocating tools18
A47L11/294… having reciprocating tools12
A47L11/26Floor-scrubbing machines, hand-driven10

Non-exclusive: a family carries several codes, including a parent and its children, so the column sums above 2,766.

Data table: Robotic-cleaning indexing subgroups
CPC Title Families
A47L2201/00Robotic cleaning machines, i.e. with automatic control of the travelling movement1,854
A47L2201/04Automatic control of the travelling movement; Automatic obstacle detection843
A47L2201/06Control of the cleaning action for autonomous devices625
A47L2201/02Docking stations; Docking operations483
A47L2201/028Refurbishing floor engaging tools, e.g. cleaning of beating brushes376
A47L2201/024Emptying dust or waste liquid containers296
A47L2201/026Refilling cleaning liquid containers249
A47L2201/022Recharging of batteries237

Non-exclusive: the parent tag A47L2201/00 sits on 1,854 families alongside one or more of its subgroups.

Inside the scrubbing branch the weight sits on the general motor-driven groups rather than on the groups that name the dirty-liquid pick-up: 914 families carry A47L11/28 and 748 its rotary-tool subgroup A47L11/282, against 118 on A47L11/29 — the group whose title states the liquid recovery that defines a scrubber-dryer — and 344 on A47L11/302, its rotary-tool-with-suction subgroup. Inside the robotic indexing scheme the base-station side is a patenting field of its own: 483 families carry the docking-station tag A47L2201/02, 296 the emptying of waste-liquid containers, 249 the refilling of cleaning-liquid containers and 237 battery recharging.

Multi-assignment. Both tables count families, and a family carries several codes — commonly a parent group and one of its children at the same time. Column sums therefore exceed 2,766 and the rows are not parts of a whole. No pie or doughnut chart is used for either cut for that reason.

Adjacent Technologies

CPC subclasses outside A47L that the same families also carry — the technical footprint beyond the cleaning machine itself.

Data table: Adjacent CPC subclasses
CPC subclass Title Families Share of 2,766
G05DSystems for controlling or regulating non-electric variables1716.2%
Y02EReduction of greenhouse gas emissions, related to energy generation, transmission or distribution1114.0%
Y02BClimate change mitigation technologies related to buildings1063.8%
B25JManipulators; chambers provided with manipulation devices662.4%
F26BDrying solid materials or objects by removing liquid therefrom652.4%
A61LMethods or apparatus for sterilising materials or objects; disinfection612.2%
B08BCleaning in general; prevention of fouling in general572.1%
D06FLaundering, drying, ironing, pressing or folding textile articles421.5%
G01SRadio direction-finding; radio navigation; determining distance or speed381.4%
A46BBrushes351.3%

Subclasses with at least 30 families. Non-exclusive: a family carries several subclasses, so the rows are not parts of a whole.

Navigation and control reach the classification record thinly: 171 families (6.2 %) carry a G05D control code, 66 a B25J manipulator code and 38 a G01S navigation code. The autonomy of these machines is recorded almost entirely through the A47L2201 tag rather than through the robotics and navigation subclasses. Two Y-section tags follow closely behind — Y02E with 111 families and Y02B with 106 — which is the battery and building-energy side of the machines being tagged, not a separate technology.

Y sections are CPC-only. Y02B and Y02E are tagging schemes with no IPC equivalent, applied across the whole scheme rather than within one technical field. A family carrying Y02E also sits in its ordinary subclass, and offices that do not assign CPC are absent from these two rows entirely.

Geographic Analysis

Which offices the 2,766 families reach, and how that footprint moves between the two halves of the window.

Data table: Top filing offices
Office Families Share of 2,766
CN — China2,56492.7%
WO — PCT50318.2%
US — United States40114.5%
EP — European Patent Office33312.0%
KR — Korea1676.0%
JP — Japan1194.3%
DE — Germany1174.2%
AU — Australia983.5%
TW — Taiwan883.2%
CA — Canada361.3%
FR — France230.8%
GB — United Kingdom220.8%
ES — Spain220.8%

26 offices carry at least one family; the 13 further offices hold fewer than 10 families each. Non-exclusive: a family reaching several offices appears in several rows.

Data table: Office footprint by period
Office 2019–2021 families Share of 1,160 2022–2024 families Share of 1,606
CN1,07092.2%1,49493.0%
WO26122.5%24215.1%
US25321.8%1489.2%
EP16714.4%16610.3%
KR998.5%684.2%
JP695.9%503.1%
AU534.6%452.8%
DE393.4%784.9%
TW363.1%523.2%
CA302.6%60.4%

Periods are defined by the family's earliest filing year: 1,160 families in 2019–2021 and 1,606 in 2022–2024. Second-period figures are affected by the publication delay for later family members.

2,564 of 2,766 families — 92.7 % — have a member at the Chinese office, and the Chinese share is the one figure that does not move between the two halves of the window (92.2 % to 93.0 %). Everything international moves down: the US share falls from 21.8 % to 9.2 % and the PCT share from 22.5 % to 15.1 %. The German office is the exception, rising from 3.4 % to 4.9 %. Part of the international decline is mechanical — a national-phase entry follows its priority filing by up to 30 months, so the 2022–2024 cohorts have had less time to acquire foreign members.

Applicant origin

Applicant country is recorded for a minority of the harmonised applicant records in this scope, so the counts below are a lower bound and are given as a table rather than a chart.

Applicant country Families 2019–2021 2022–2024
CN — China427206221
DE — Germany853154
KR — Korea836716
US — United States764432
TW — Taiwan1376
CA — Canada1248
NL — Netherlands1028
FR — France725
GB — United Kingdom716
DK — Denmark514

Coverage note. Of the 2,764 scope families that carry at least one applicant record, only 731 (26.4 %) carry a non-blank applicant country code. The country table is therefore a lower bound and cannot be read as a distribution. The gap is widest for China: 2,564 families have a member at the Chinese office while 427 carry a Chinese applicant country code.

Applicant Landscape

Applicants inside the scope, with PATSTAT han_name variants consolidated by name string.

PATSTAT's harmonised applicant name splits several companies across spellings — among the applicants with at least four families, Roborock appears under four han_name forms, Silver Star under four and Yunjing under seven. The counts below consolidate those variants by matching on the distinctive part of the name. The consolidation is deliberately conservative: it merges spellings of the same name and nothing else, so entities that file under a different brand name remain separate rows.

Data table: Applicants with the most families in scope
Applicant Type Families Share of 2,766 With CN member With WO member With EP member
Beijing RoborockCompany1786.4%12613280
Shenzhen Silver StarCompany1696.1%166312
Dreame InnovationCompany1415.1%1333011
Gree ElectricCompany1023.7%10200
Yunjing (Narwal)Company752.7%712219
EcovacsCompany692.5%6862
MideaCompany672.4%52246
Meizhi ZonghengCompany622.2%6200
LG ElectronicsCompany612.2%244235
Beijing StoneCompany572.1%5700
TinecoCompany531.9%5241
HaierCompany521.9%5210
SharkNinjaCompany511.8%4683
PositecCompany451.6%452013
3irobotixCompany351.3%3500

Non-exclusive: a co-filed family counts for each of its applicants, so the column sums (1,217 across these fifteen rows) exceed the number of distinct families they cover.

No applicant holds more than 6.4 % of the scope: the largest count is Beijing Roborock with 178 families, followed by Shenzhen Silver Star with 169 and Dreame Innovation with 141. What separates them is not volume but reach. Roborock takes 132 of its 178 families through the PCT route and 80 to the EPO; Gree Electric, Meizhi Zongheng, Beijing Stone and 3irobotix take none of theirs to either. LG Electronics files the smallest home share of the group — 24 of 61 families reach the Chinese office, 42 the PCT route and 35 the EPO.

Consolidation limits. Name-string consolidation cannot see corporate ownership. Where a group files through subsidiaries carrying unrelated names, those subsidiaries appear as separate rows and each row understates the group. The rows above are applicant entities as PATSTAT harmonises and this report merges them — they are not corporate groups.

Professional Segment

The manufacturers of professional cleaning equipment and commercial cleaning robots inside the scope, identified by applicant name.

Classification records what a machine is, never who buys it. There is no CPC or IPC code that separates a scrubber-dryer sold into a supermarket from one sold into a kitchen, so the commercial dimension of this subject cannot be read out of the patent classification at all. This section therefore adds an explicit interpretive layer: fourteen applicants whose business is professional cleaning equipment or commercial cleaning robots were assigned by hand, and every family carrying one of them was counted. The rule and its limits are stated in the Methodology section; the assignment is an editorial judgement about companies, not a property of the patent data.

* 2023–2024: incomplete data due to the ~18-month publication delay (PATSTAT Global, Spring 2026, patstat_2026a).

Data table: Professional-segment families by year
Year Professional-segment families All families in scope Professional share
2019111776.2%
2020173704.6%
2021216133.4%
2022196203.1%
2023*155612.7%
2024*124252.8%
Total952,7663.4%

Each family is counted once, in its earliest filing year; the yearly figures sum to 95. No family in this cohort carries two of the fourteen applicants.

The fourteen applicants

Applicant Type Families Granted With WO member With EP member With CN member
Gaussian Automation (Shanghai)Company28150028
Alfred KärcherCompany21615109
AvidbotsCompany83311
HakoCompany75075
Pudu TechnologyCompany74107
Ecovacs Commercial RobotCompany53005
NilfiskCompany51220
Suzhou Chenghe Cleaning EquipmentCompany51005
Kingsun Cleaning Equipment (Suzhou)Company42004
Adlatus RoboticsCompany10000
Brain CorpCompany10100
LionsBot InternationalCompany11110
TennantCompany11111
WetrokCompany10100
Cohort total9542252265

The professional cohort holds 95 of 2,766 families, 3.4 % of the scope, and its annual output stayed inside a band of 11 to 21 families across all six years. Because the surrounding volume grew from 177 to 620 families a year, the cohort's share of the field fell from 6.2 % in 2019 to 3.1 % in 2022 without its own output falling. Two applicants account for half of it: Gaussian Automation with 28 families and Alfred Kärcher with 21; the other twelve hold eight or fewer each. Their filing routes point in opposite directions — all 28 Gaussian families reach the Chinese office and none the PCT route or the EPO, while 15 of Kärcher's 21 take the PCT route and 10 reach the EPO.

Small numbers. Twelve of the fourteen applicants hold eight families or fewer, five of them a single family. At those counts a difference of one or two families is not a signal, and no ranking, trend or growth statement is made about individual companies below Gaussian Automation and Alfred Kärcher. A single-family count also means the applicant filed once in this precise classification intersection, not that it holds one patent family overall.

Collaboration

Families filed by more than one applicant, and the applicant pairs that appear on the same application.

Co-applicant pairs are read from the application, not from the family. Pairing applicants that appear anywhere in the same family would match a company with its own name variants filed at different offices and report that as collaboration; joining on appln_id keeps a pair to what two applicants actually signed together.

* 2023–2024: incomplete data due to the ~18-month publication delay (PATSTAT Global, Spring 2026, patstat_2026a).

Data table: Co-filed families by year
Year Co-filed families All families Co-application rate
20192317713.0%
20203937010.5%
2021376136.0%
2022606209.7%
2023*8956115.9%
2024*284256.6%
Total2762,76610.0%

A family counts as co-filed when at least one of its members carries more than one applicant. Rate = co-filed families / all families of the same filing year.

Applicant pairs

Applicant A Applicant B Joint families
Guangzhou Shirui Electrical TechnologyGuangzhou Shiyuan Electronic Technology9
Chen Xiaoping IndividualFoshan Viomi Electric Appliance Technology7
Haier Smart HomeQingdao Taboer Robot Technology6
Shenzhen Youjiyuan TechnologyUBTech Robotics6
Guangzhou CVTE Electronic TechnologyGuangzhou Shirui Electrical Technology6
Kingclean ElectricLexy Electric Green Energy Technology (Suzhou)4
Gree Electric Appliances of ZhuhaiZhuhai Leayun Technology3 + 2 †
Jiangsu Laike Intelligent Electric ApplianceKingclean Electric3
Jiangsu Laike Intelligent Electric ApplianceLexy Electric Green Energy Technology (Suzhou)3
Shenzhen Youbiye Science and TechnologyShenzhen Youyuan Science and Technology3
Shenzhen Jingchuang Zhihang TechnologyXiaowo Intelligent (Beijing) Technology3
Shanghai Gaussian Automation TechnologySuzhou Gaozhixian Automation Technology2
Shenzhen Beatles IntelligentHangzhou Joyoung Household Electrical Appliances2
Beijing Shunzao TechnologySuzhou Xiaoshun Technology2
Chen, Simon Nai Pong IndividualIntelligent Cleaning Equipment Holdings2

Pairs with at least two joint families, after the exclusions noted below; four further qualifying pairs with two or three joint families are not listed. † Two rows in the query result, split by two han_name spellings of Gree Electric; the two sets of families are disjoint, so the pair covers 5 families in total.

276 of 2,766 families carry more than one applicant — 10.0 %, and no collaborating pair reaches double digits. The largest is Guangzhou Shirui with Guangzhou Shiyuan at 9 joint families. Among the pairs with at least two joint families, one comes from the professional segment: Shanghai Gaussian Automation with Suzhou Gaozhixian Automation, 2 joint families. Pairs that share only a single family are below the query's threshold and are not visible here. The pattern that recurs is domestic and supplier-shaped — appliance brands filing with a technology supplier in the same city, or a named individual filing alongside a company. The 2023 figure of 89 co-filed families sits inside the provisional window and is not interpreted as a change in co-filing behaviour.

Self-inclusion check and its residue. Applicant names were normalised before pairing, so no company is paired with its own han_name spellings. Two pairs nevertheless survived that are affiliated rather than collaborating, because the two entities carry genuinely different names: Narwel with Yunjing (16 families) and Cloud Whale with Yun Whale (2 families, two translations of the same Chinese name). Both are excluded from the table above. Comparable cases that this report has not identified may remain.

Grant Rates

Share of families with at least one granted member, by the cohort's earliest filing year.

* 2022–2024: grant rates for these cohorts are artificially low because examination is still pending for much of them. The headline rate quoted in this report uses the 2019–2021 cohorts only.

Data table: Grant rate by filing cohort
Cohort Families With a granted member Grant rate
201917712168.4%
202037022059.5%
202161332252.5%
2022*62024639.7%
2023*5619617.1%
2024*4257517.6%
2019–2021 (completed cohorts)1,16066357.2%
All years2,7661,08039.0%

What is being granted

IPR type of the family member Families containing such a member With a granted member With a CN member
PI — patent of invention2,7601,0132,513
UM — utility model124124102
DP — design patent110

Non-exclusive: a family holding both a patent of invention and a utility model appears in two rows, so the rows sum above 2,766. Design filings were excluded at the application level (appln_kind != 'D'); one family retains a member typed as a design patent under a different application kind.

For the three cohorts old enough to have completed examination, 663 of 1,160 families hold at least one granted member — 57.2 %. The decline across 2022 to 2024 is pendency, not refusal: those families are largely still in examination, and the 2024 figure of 17.6 % will rise with every further PATSTAT edition.

Utility models are inside the scope. 124 families contain a utility-model member, and a utility model registers as granted without substantive examination. The grant rates above therefore mix two different acts. Those 124 families are 4.5 % of the scope; the report keeps them and states their number rather than filtering them out silently.

Methodology

Search strategy, data basis, counting rules and known limitations.

Data basis. This report was produced on 2026-08-15 on EPO PATSTAT Global, Spring 2026 (patstat-mtc.patstat_2026a), with the classification scheme editions IPC 20260101 and CPC 202608 in force. The classification set was completed in three steps before any family was counted: the three seed groups were expanded to their full subtree by a recursive walk on parent (3 codes → 13, and A47L2201/00 → 8), classification predecessors were unioned from the reclassification chain (no predecessor entry exists for either group), and retired symbols were kept unfiltered (neither group contains a retired symbol in IPC or CPC). Every figure in this report comes from a query run against that edition.

Search strategy, family counting, multi-assignment and data source

Search strategy

  • Three entry points into the classification scheme were used: the WIPO catchword index (dpt_catchword), the title breadcrumb (SEARCH(title_full, …) on dpt_cpc, singular and plural forms searched separately and unioned) and the CPC definition texts (SEARCH(definition_scope, …)).
  • The catchword index maps „SCRUBBING of floors“ to A47L11/26, A47L11/28 and A47L13/10. The third is a hand implement and was excluded, as was B01D47/00 („SCRUBBING of gases“), the homonym.
  • Layer 1 is the floor-scrubbing subtree: A47L11/26, /28, /282, /283, /284, /29, /292, /293, /294, /30, /302, /305, /307.
  • The autonomy condition is the CPC indexing scheme A47L2201 (8 symbols). A family enters the scope when it carries at least one code from each set. The recursive subtree of A47L2201/00 and the prefix A47L2201% were confirmed to return the same 8 symbols.
  • Layer 2 was checked and produced nothing additional: the CPC definition texts return A47L11/00 and A47L23/00 only, and no application-domain tagging scheme outside A47L names floor scrubbing. Y-section tags do appear on the resulting families (Y02E 111, Y02B 106) but as a consequence of the machines, not as an entry point.
  • Layer 3, a title search, was run as the negative test rather than as a widener — see Scope Limitations.
  • Classification system: CPC throughout. This is not a preference here but a constraint: the IPC scheme contains A47L11 but has no A47L2201 equivalent, so there is no IPC route to the autonomy condition.

Family counting

  • All counts are DOCDB patent families (docdb_family_id); pseudo-families (docdb_family_id ≤ 0) are excluded.
  • A family is assigned to the earliest filing year across all its members, and the scope keeps families whose earliest year falls in 2019–2024. Each family therefore counts in exactly one year and the yearly figures sum to 2,766 exactly.
  • Design filings are excluded at application level (appln_kind != 'D'). Utility models are not excluded; 124 families contain a utility-model member.
  • Office, PCT and grant figures count every non-design member of a scope family, including members filed outside the 2019–2024 window.
  • Co-applicant pairs are joined on appln_id, never on docdb_family_id: a family-level join pairs a company with its own name variants filed at different offices.
  • Applicant name variants are consolidated by matching the distinctive part of the han_name string. The consolidation merges spellings of one name and does not follow corporate ownership.

Multi-assignment

  • A family carries several CPC codes, frequently a parent group and its child at the same time. Every classification table and chart in this report is therefore non-exclusive and its column sums exceed 2,766.
  • For the same reason no pie or doughnut chart is used anywhere in this report: those forms imply mutually exclusive parts of a whole.
  • A co-filed family counts for each of its applicants, so applicant columns also sum above the number of distinct families covered.

The professional segment

  • Neither CPC nor IPC distinguishes a machine sold into a commercial building from one sold into a household. The professional segment is therefore an editorial assignment, not a data property.
  • Rule applied: an applicant is counted as professional when its business is professional or industrial cleaning equipment, or commercial cleaning robots. Fourteen applicants qualified: Alfred Kärcher, Nilfisk, Tennant, Hako, Wetrok, Avidbots, LionsBot, Adlatus Robotics, Brain Corp, Gaussian Automation, Pudu Technology, Ecovacs Commercial Robot, Kingsun Cleaning Equipment and Suzhou Chenghe Cleaning Equipment.
  • Applicants whose market position spans both segments were left out of the professional cohort rather than assigned by guess. The cohort is a floor, not a census.

Data source

  • EPO PATSTAT Global, Spring 2026 (patstat-mtc.patstat_2026a) on Google BigQuery, queried through the patstat-mcp server.
  • Classification views patstat-mtc.classification.dpt_cpc, dpt_ipc, dpt_catchword, dpt_ipc_reclassification, dpt_cpc_reclassification and dpt_edition.
  • Core tables: tls201_appln, tls202_appln_title, tls206_person, tls207_pers_appln, tls209_appln_ipc, tls225_docdb_fam_cpc.
  • All SQL is reproduced in queries.sql alongside its results.

Scope Limitations

  • Title-only families: 562. A title search for a scrubbing term combined with an autonomy term returns 605 families whose earliest filing falls in the window; 562 of them, 20.3 % of the scope, do not carry the classification profile. The scope was kept tight rather than widened: the sampled profile is clean at 0 off-topic in 22, and widening it to absorb those families would trade that away. The 562 are a known gap, not an argument that the figures are wrong.
  • The autonomy marker is CPC-only. 18,696 families carrying an IPC floor-scrubbing code have their earliest filing in this window, and 10,669 of them — 57.1 % — carry no CPC assignment at all. Those families cannot be reached by A47L2201 under any formulation of the query. Offices that do not assign CPC are structurally under-represented in every figure in this report.
  • Applicant country is 26.4 % covered. Only 731 of the 2,764 scope families with an applicant record carry a non-blank applicant country code. The applicant origin table is a lower bound and is not a distribution.
  • Utility models are included. 124 families contain a utility-model member, and a utility model registers as granted without substantive examination.
  • Recent years are provisional. The ~18-month publication delay under-counts the 2023 and 2024 cohorts, and it under-counts later members of every recent family. No growth or decline statement in this report rests on 2023 or 2024.
  • Grant rates for 2022 onwards are pendency, not outcome. The headline grant rate uses the 2019–2021 cohorts only.
  • The professional cohort may under-count. It rests on recognising company names. An applicant selling into the commercial market under a name this report did not recognise is counted outside the cohort.
  • External plausibility was not tested against a second database. No Lens.org, Espacenet or Google Patents comparison was run for this scope in this session. The internal consistency checks that were run are listed in queries.sql.
Glossary — patent terms explained
DOCDB patent family
A group of patent applications protecting the same invention in different countries. Counting families instead of applications keeps one invention filed in five countries from counting five times.
Earliest filing year
The lowest filing year across all members of a family. Used here so that each family appears in exactly one year of the trend.
IPC / CPC
International and Cooperative Patent Classification — hierarchical schemes for classifying patents by technology. CPC is maintained jointly by the EPO and the USPTO and is applied consistently across the members of a family; IPC is assigned by almost every office.
Indexing scheme (the 2000-series)
CPC groups numbered 2000 and above, such as A47L2201, are tags rather than places: they qualify a document already classified elsewhere. They exist in CPC only.
Scrubber-dryer
A floor-cleaning machine that applies cleaning liquid, scrubs the floor and takes the dirty liquid back up in one pass. In CPC this is named by A47L11/29 and its subgroups.
PCT / WO application
An international application under the Patent Cooperation Treaty. It preserves the option of protection in many countries; the national decisions follow later.
Utility model
A registered right for a technical invention, granted in some jurisdictions without substantive examination and with a shorter term than a patent.
Publication delay
Applications are usually published about 18 months after filing, so the most recent filing years are always incomplete in any patent database.

Report Files

The complete SQL of this report, including the discovery queries, the frozen scope CTE, the quality-assurance checks and the result of every query, is stored alongside the report as queries.sql. Re-running it against a later PATSTAT edition or a later classification edition will not reproduce these figures exactly, which is why both editions are named at the top of this section and in the file header.

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. Every number in it comes from a query stored in queries.sql.

Custom Patent Report Need a similar analysis for your technology area, company, or research question? We build tailored patent intelligence reports.
The Toolchain Want to run your own AI-powered patent analyses? We help you set up PATSTAT on BigQuery with MCP server integration.

Data: EPO PATSTAT Global, Spring 2026 (patstat_2026a) · mtc.berlin