Hospital Cleaning: What Robots Take Over — and What Stays With Trained Staff

Three terms decide every hospital cleaning tender, and they are not synonyms. KRINKO defines surface cleaning as a process that removes soiling using water with cleaning-enhancing additives such as surfactants; micro-organisms are removed mechanically along with it, "without a kill or inactivation taking place or being intended as designed". Disinfection, by contrast, is defined as a process that reduces the number of viable micro-organisms "with a standardised, quantifiable proof of efficacy". Disinfecting surface cleaning combines both in a single work step. And routine maintenance cleaning — the term German tenders call Unterhaltsreinigung — is defined in the recommendation as cleaning within basic hygiene. A tender that blurs these three ends up with the wrong machine and the wrong expectations.

The second question is who does the work, and the recommendation is unusually concrete about it. Where an external service provider is engaged, the scope of cleaning (maintenance cleaning plus additional cleaning services) and of surface disinfection is to be compiled in a specification of services, broken down by risk area and by contamination risk, room by room and surface by surface. Interfaces with the provider and the tasks that the provider does not perform are to be fixed in the cleaning and disinfection plan. That turns the robot question into a procurement question: not "does the machine clean well", but "which lines of the specification does it actually cover, and who covers the rest".

The third factor is the operating window, and it is what makes hospitals different from retail. A hospital never empties out. Corridors carry traffic around the clock, wards have their own morning routine, functional and operating units run to their own rhythms. So the robot window is not "the night" in the sense a supermarket means it — it is the low-traffic band plus the daytime public zones where a machine can work alongside people. Which machines fit which zones on your site is what our hospitals industry page sets out, and the rest of this article explains the reasoning behind it.

KRINKO has a section on machine surface cleaning, and its recommendation is one sentence long: the commission recommends "using machine methods for surface cleaning only in areas without infection risk, for lack of efficacy evidence" (category II, i.e. based on indicative studies and rigorous, plausible theoretical derivation). The reasoning sits in the body text. Scrubber-dryer machines can be used on plastic floor coverings and tiles — but there are currently no test procedures for demonstrating the cleaning or disinfecting efficacy of machine methods. On scrubber-dryers specifically, the recommendation notes that their use for floor cleaning in healthcare facilities did not lead to increased microbial air contamination when properly operated and maintained; but because no data exist on the efficacy of disinfecting surface cleaning by machine, "these machines should be used for cleaning purposes only".

That comes with a condition almost nobody asks about in a tender. Operating the machines according to the manufacturer’s specifications for maintenance and reprocessing is stated as a precondition — with hoses, squeegee lips and brushes that can be reprocessed machine-(chemo)thermally being preferred — and the recommendation adds that these preconditions and the reprocessing cycle "are to be evidenced by the equipment manufacturers by way of expert opinion". That is a question you can put to any vendor, ourselves included, and it is asked far too rarely. None of this is law: KRINKO publishes recommendations. But § 23 (3) sentence 2 of the German Infection Protection Act presumes compliance with the state of medical science where the published KRINKO recommendations have been observed, which is why in practice they set the benchmark.

Read the right way, this is the argument for robots, not against them. The disinfecting wipe of frequently touched, patient-near surfaces is qualified manual work that a floor machine was never going to do. Every hour the floor stops consuming is an hour available for exactly that work — plus for sanitary areas, edges and corners, the space under beds, isolation rooms and terminal disinfection, all of which stay with your team. Robots take the open floor, not the occupation, which is the same conclusion our piece on the staff shortage in commercial cleaning reaches from the labour-market side.

The recommendation sorts rooms into risk areas and assigns measures to each. Areas without elevated infection risk are listed as stairwells, corridors, administration, offices, dining rooms, lecture halls, teaching rooms, technical areas and waiting rooms; for all surfaces there, the measure is cleaning. In areas with possible infection risk — general wards, outpatient areas, radiology, physical therapy, sanitary areas, dialysis, delivery rooms, functional diagnostics, psychiatry, procedure rooms, ambulance and patient transport vehicles, waiting rooms — frequently touched and patient-near surfaces and barefoot areas require disinfecting surface cleaning or surface disinfection, while floors and rarely touched surfaces require cleaning. In areas with elevated infection risk (operating units, intensive care and IMC, severe burns, transplantation, haemato-oncology, neonatal intensive care) and in areas with special infection risk such as isolation, the floors themselves require disinfecting surface cleaning.

That produces a tension worth stating rather than smoothing over. On a general ward the floor is, per the risk table, a cleaning surface — which reads like robot territory. But the machine-specific recommendation limits machine surface cleaning to areas without infection risk. We do not resolve that contradiction, and no vendor should: the assignment of rooms to risk areas and of methods to surfaces is made facility-specifically by the hospital hygiene team in the cleaning and disinfection plan, which is where it is laid down for every surface when, with what, and how it is to be cleaned or disinfected. What we can say is where the machine is uncontroversial — and that zone is large.

One detail shows why these lists have to be walked through per building instead of copied. Waiting rooms appear in two columns of the same table: in the no-elevated-risk column only where patients with suspected infection and patients with pronounced immunosuppression are separated, and otherwise among the areas with possible infection risk. Whether the entrance hall of your outpatient department is robot territory is therefore not a machine question at all. It is a question about your patient flow, and the hygiene team answers it.

The corridor machine in our fleet is the Pudu CC1. Per Pudu’s official operating manual V4.2 it works at 700–1,000 m²/h with a 500 mm cleaning width including the side brush, carries 15 litres of fresh and 15 litres of waste water, runs 5 hours in scrubbing mode and up to 9 hours in silent mopping, needs a minimum path clearance of 70 cm, and is specified at below 70 dB operating noise. Those 70 cm are the number that decides hospital deployments: doorways, bed bays and corridor constrictions are the geometry that either lets a machine through or does not. There is also the CC1 Pro, the same platform with an added AI layer, which we come back to in the documentation section — one caveat belongs here already: it is not the quieter machine, and on that figure Pudu contradicts itself — the May 2025 product deck says below 70 dB, the current official German product page says below 75 dB.

The floor covering deserves an honest look, because hospitals are the vertical where the official application lists get uncomfortable. Pudu’s manual gives the CC1’s application scope as terrazzo, marble, tiles, epoxy resin, sandstone, artificial stone and low-pile carpet, followed by "etc." — linoleum and PVC, the two most common hospital floors, do not appear on the list, though the "etc." means the list is not exhaustive. KRINKO, for its part, names scrubber-dryer machines explicitly for plastic floor coverings and tiles. Those two statements do not contradict each other, but they do not add up to a blanket yes either. We clear the covering per surface, with the manufacturer, before a machine is scheduled onto it — the same rule our scrubber-dryer guide applies to wood and laminate.

The CenoBots L3 is the alternative, though not because it covers more ground. Its cleaning width is the narrower of the two — 400 mm against the CC1’s 500 mm — and the up to 2,016 m²/h the official 2026 catalogue prints is labelled maximum theoretical productivity, which is not a rate any ward gets planned from. What it offers instead is running longer without anyone attending it: 25-litre solution and 25-litre recovery tanks against 15 and 15, up to 4 hours runtime (ECO mode) and 1.5 to 2 hours charging, and a workstation that refills, drains and charges in one place. At 700 mm minimum passable width it still reaches the same tight geometry, at 127 kg. CenoBots lists hospitals among the machine’s commercial segments; the DeepClean mode is a platform feature the catalogue describes for the CenoBots line as a whole, and it is in that section — not in the L3 data block — that hospitals are named, alongside sports halls, schools and restaurants. We publish no decibel figure for the L3: the catalogue prints "≥ 64 dB(A)", and the available sources contradict each other across a range of roughly 60 to 66 dB(A) — we would rather have no number than a chosen one.

The third machine is the odd one out, and in a hospital that is exactly its point. The Nexaro NR 1700 is a dry vacuum robot whose official datasheet lists a long-range laser distance sensor, a bumper sensor, eight drop sensors, two magnetic field sensors, a wall-following sensor, incremental sensors and an inertial measurement unit — and no camera. It is documented at 49 dB(A) in Eco mode, 280 mm cleaning width, up to 1,000 m² per job and 250 minutes of Eco runtime. The datasheet’s 100 m²/h is measured on carpet; in our own deployments the effective rate is 50 to 75 m²/h and 200 to 300 m² per charge. That makes it the machine for the administration wing, offices and small carpeted waiting areas — not for corridor kilometres.

The documentation duty in a hospital is not a matter of taste. Under § 23 (5) of the Infection Protection Act, the heads of hospitals and comparable facilities must ensure that internal procedures for infection hygiene are laid down in hygiene plans. As part of that plan, KRINKO expects the cleaning and disinfection plan to specify, for all surfaces, when, with what and how they are to be cleaned or disinfected, including instructions on when a surface may be used again, with indications, target objects and repeat intervals derived from the risk area and the contamination risk. That is the document a robot has to fit into. It is not a document a robot produces.

What the machines do contribute is process evidence. A cleaning run leaves a record: which area was driven, when it started and ended, in which mode. The CC1 Pro goes one step further — Pudu describes a rear-facing AI camera monitoring cleaning performance in real time, detecting remaining soil, triggering spot follow-ups and feeding heatmaps that mark heavily soiled areas and stubborn stains, alongside an operations dashboard with task completion and maintenance data. Its AI spot mode is rated at 1,500 to 3,000 m²/h, which is a patrol rate and not a full-coverage rate; the difference is the subject of our CC1 versus CC1 Pro comparison. For a hygiene audit that is genuinely useful material, because it is machine-generated and does not depend on someone ticking a list at the end of a shift.

And here is the boundary, stated plainly because the industry keeps blurring it. KRINKO expects facility-specific hygiene monitoring to assure the quality of surface cleaning and disinfection, and the methods it tabulates are visual assessment as a checklist audit, swabs, contact cultures, fluorescence gel and ATP systems — a robot report is none of them. The recommendation even notes that for cleaning as such there is still no valid test method to quantify it, and its own note on the ATP method is that it measures cleanliness at the moment of testing but not the cleaning process. Which is exactly the division of labour: the robot documents the process, the hygiene monitoring assesses the result. A coverage heatmap is supporting evidence in an audit, never a hygiene certificate, and anyone selling it to you as compliance is overselling. How that evidence is structured across verticals is the subject of our piece on cleaning verification.

Two more requirements land on the robot rather than on the floor. Where an external service provider is engaged, KRINKO expects it to define procedural instructions in agreement with the hygiene team, explicitly "including the handling of special methods/technologies" — an autonomous machine is such a technology, and its handling belongs in writing, not in a WhatsApp group. And before new technologies are introduced, the recommendation states that additional training is required. Budget the training. It is the cheapest line in the project and the one that decides whether the machine still runs in month four.

Night in a hospital means low traffic, not an empty building, and that changes what the noise figures mean. Below 70 dB is a perfectly reasonable value for a scrubber-dryer working a corridor, an entrance hall or a canteen at four in the morning. It is not a value for a machine passing patient room doors while people sleep behind them, and no marketing phrase makes it one. The quiet machine in this comparison is the vacuum at 49 dB(A) in Eco mode, and it is quiet because it is slow. What unattended operation additionally requires — charging logic, a route by which a three-a.m. fault reaches a human, doors, lifts and the alarm system — is worked through in our article on night operation. In a hospital, fire doors and lift access are usually the harder half of that list.

The camera question decides more hospital projects than the cleaning performance does. Our fleet is not camera-free: Pudu’s official CC1 brochure lists two RGBD cameras and a top-view camera alongside two lidars, and the CenoBots catalogue documents a depth camera for the L3. The Nexaro NR 1700 is the machine whose published sensor list contains no imaging device at all. Where a data protection officer or a works council draws a hard line on cameras in patient areas, the honest answer is machine choice rather than reassurance — and the questions to settle in writing before signing, from the processing agreement to the storage region, are collected in our piece on cleaning robots and GDPR. None of this is legal advice; the assessment of your specific deployment belongs to your data protection officer.

One honest note on scale, because hospital campuses can outgrow everything described here. For very large connected areas there is a camera-free German machine available through our programme, newly onboarded there: per its own documentation the Adlatus CR700 navigates without cameras, records no personal data but only coordinates, and pairs that with certified cleaning documentation — an argument that lands well in public procurement. Adlatus’s own phrasing, that it is the only manufacturer navigating without high-resolution cameras, we do not repeat: the NR 1700 in our own fleet is camera-free too. Our assessment of the machine is documentation-based, and it is written up in our Adlatus CR700 market overview. Which class fits your building at all, before any brand question, is what our guide on choosing a cleaning robot walks through.