Cleaning Parking Garages: Which Machine Actually Works on a Park Deck

The base load in a parking garage is abrasion. Tyre rubber and brake dust settle as a fine, dark, slightly greasy film that binds to the floor and, once it is walked and driven into the surface, no longer comes off dry. On top of that sits everything loose: grit and salt residue carried in on tyres through the winter, leaves and paper blown through the open sides, cigarette butts at the lift lobbies, drinks packaging and receipts around the payment machines, and the occasional oil drop under a parked car. It is a debris profile that matches what the sweeper manufacturers describe almost word for word — Pudu's official MT1 brochure names "larger debris like leaves and bottles", and the MT1 Max deck says the machine removes "everything from cigarette butts and paper scraps to large bottles in a single sweep".

Geometrically, a parking structure is simple; operationally, it is difficult. The driving lanes are long, wide and largely free of obstacles, which is close to ideal for a machine. The bays are occupied in a pattern that changes hourly. The ramps concentrate everything: every vehicle carries water, salt and dirt over the same few metres. And the floors are not one floor — the deck itself is bare or coated concrete, while the pedestrian zones around lifts, stairwells and payment machines are usually tiled or coated. It also matters a great deal whether you are dealing with a fully enclosed underground car park, a parking garage with open sides, or a fully exposed top deck: the first has no weather but concentrated dust and poor ventilation, the second has wind-blown debris and damp entry zones, the third is outdoor in every practical sense.

Who cleans it, and how often, is usually the weakest part of the arrangement. Parking structures are the low-priority area in most cleaning contracts — a few machine passes a year plus a periodic deep clean — because they compete for the same scarce staff as the areas customers actually look at. That competition is the real driver behind automation here, and we set it out in our piece on the staffing shortage in commercial cleaning. The windows follow occupancy rather than convention: an office garage empties at night and on weekends, a shopping-centre garage after closing, and a residential garage is fullest at night and emptiest in mid-morning — the opposite of the usual night-cleaning assumption.

A parking structure applies three gates to every machine, and all three are written in the operating manuals rather than on the datasheets. First: is the machine released for anything other than an enclosed indoor room, and does it carry a protection rating at all? Second: what is its stated operating temperature band, given that a deck near the entrance sits close to outdoor temperature in January? Third: what gradient may it climb in automatic mode — which is the only mode that matters for an unattended run. Most disappointments in this vertical are one of those three lines, not a throughput problem.

Take the base Pudu MT1 first, because it is the machine people assume will do this job. Officially it is a strong sweeper: max. 1,800 m²/h in standard mode and max. 6,000 m²/h in spot mode, 70 cm cleaning width, a 35-litre bin, 45 Ah, 4–8 hours of run-time differentiated by cleaning level, under 3 hours charging, a minimum path clearance of 75 cm, and it crosses 20 mm thresholds and 35 mm grooves. Pudu calls it "professional-grade dry cleaning". What its official material contains nowhere is an IP rating or a single outdoor scenario — and its user manual restricts the machine to indoor flat environments, forbids use outdoors on rough or non-hardened surfaces, and sets 0 to 40 °C with at most 90 % relative humidity. In practice that limits the MT1 to a fully enclosed garage that stays reliably above freezing, and even there it has no published protection against a wet floor.

The wet class does not get to the deck at all, and it is worth saying why rather than leaving it off a list. The CenoBots L50 is our largest scrubber, and three of its own official constraints hit simultaneously. Its user manual lists among the situations in which the machine must not be operated: "Outdoors. This machine is for indoor use only." No IP protection rating is published for it anywhere — not in the manual, not in the datasheet, not on the product page — so no promise about spray or driving rain can be made. And the same safety list caps autonomous operation at "a gradient exceeding 2%", with 10 % applying only to manual operation with a person walking the machine. Even in a fully enclosed underground car park, a ramp exceeds 2 % many times over. We go through the machine and its limits in detail in our L50 profile. The short version for this vertical: the wet class stops at the ramp.

That leaves one machine in our fleet, and it is not a compromise but the machine Pudu wrote this brief for. The official product title of the MT1 Max is "AI-Powered Outdoor & Semi-Outdoor 3D Perception Robotic Sweeper". Its product deck opens with the sentence: "Building on MT1, MT1 Max is enhanced for semi-outdoor environments such as residential courtyards and for areas with high-speed dynamic obstacles like parking garages." The launch release places it in "environments such as underground parking garages and semi-open building atriums". And in the deck's own scenario gallery, "Parking Garage" is the first tile, with six of the twenty-four scenarios shown being parking areas of one kind or another. How the two sweepers differ beyond that is the subject of our MT1 versus MT1 Max comparison, and the wider question of what "semi-outdoor" means for a sweeper is covered in our piece on sweeping robots outdoors. The machines we run every day are not the whole programme, though. Three more are rated suitable for parking structures — the sweepers CenoBots S5 and Adlatus SR1300, and the scrubber Adlatus CR700, for which parking is an official application field. All three are available through us but newly onboarded, so we assess them from documentation rather than from operating hours of our own.

Start with the work itself. The MT1 Max is a pure dry sweeper: its side brushes and main roller brush move debris into the hopper, a filtered airflow controls the dust, and there is no water anywhere in the process — the deck is driveable the second the machine has passed, which on a surface that is never fully closed to traffic is not a detail but the whole reason to choose a sweeper. Officially it runs at max. 2,200 m²/h in standard mode and max. 7,000 m²/h in spot mode, with a 70 cm cleaning width including the side brush, zero-gap edge cleaning, and a 35-litre waste container. Battery capacity is 60 Ah, run-time is stated as 5–10 hours with standard mode explicitly given as 8 hours, and charging takes under 3.5 hours. One caveat that applies to the whole family: do not multiply the top area rate by the top run-time. Pudu states the run-time as a band, and the highest cleaning rate does not sit at the same operating point as the longest run. Size the deployment from a measured trial, not from a product of two maxima.

The part that makes it legal on a semi-open deck is the environmental package, and Pudu documents it precisely. Full IP54 sealing "defends against dust ingress and water spray". A water-drop sensor "automatically detects water drops and triggers a protective avoidance mode, guiding the robot to safe zones while reducing electrical risks". The operating temperature is given as −10 to 45 °C, with a temperature sensor. The machine crosses 5 cm speed bumps and climbs slopes up to 8°. Speed bumps in particular are worth noting: they are standard equipment in parking structures, and 20 mm — the base MT1's documented threshold — does not clear one. There is also an internal vibration system that keeps the filters clear during the run, which matters more here than in a warehouse, because brake and concrete dust load a filter quickly.

The third piece is traffic, and this is where the Max carries hardware the base machine does not. Pudu's official material states: "Equipped with a 1.2-meter high warning light and a safety projection light, combined with audible and visual alarms, MT1 Max enhances visibility for pedestrians and vehicles, proactively preventing potential risks." The component diagram labels a dedicated forklift safety light and an emergency button. The behaviour is described just as concretely: the robot can "stop to let vehicles pass, pause at lane entrances, or edge aside while keeping the route clear and traffic undisturbed", and it detects sudden movement from people or animals. Its 3D LiDAR is credited with detecting objects up to 150 metres away at centimetre-level accuracy and with keeping stable mapping under glass roofs, in semi-open areas and in low light. None of that replaces on-site traffic rules, marked routes and a risk assessment — what documentation does and does not settle is the subject of our article on safety and standards, which also notes that Pudu publishes no norm citation for this particular machine.

Now the honest fine print, because it belongs in the same section. Pudu footnotes the water-drop sensor with "Do not perform cleaning tasks in extreme weather" — IP54 is protection against dust and spray, not a licence to run through a storm. Pudu publishes no noise value for the MT1 Max anywhere; the only sourced figure is a single German dealer's "< 75 dB", and the same dealer lists the base MT1 at max. 70 dB, so nothing supports the Max being the quieter machine. No manufacturer in this class publishes anything about resistance to road salt, so we make no claim there either: IP54 describes dust and water, not corrosion. And the largest limit is the category itself. This is a sweeper. It removes what lies on the deck. It does not touch what sticks to it.

The ramp is the first thing to settle, and it settles with a measurement rather than an assumption. The MT1 Max is documented for slopes up to 8°, which is a gradient of roughly 14 %. Plenty of parking ramps are steeper than that, and a ramp's gradient is not obvious by eye — the drive up feels the same at 10 % and at 16 %. So measure it before anything else, and measure it at the steepest section rather than as an average over the whole run. If the ramp stays inside the figure, the robot transits between decks on its own. If it does not, deck changes become a manual task: Pudu documents a retractable handle expressly so staff can move the robot between cleaning areas, and optional Elevator Control and E-gate Control modules for buildings with a suitable lift and gated access. For contrast, the L50 caps autonomous operation at 2 % gradient, which rules out every ramp in every parking structure without exception. Either way, deck-to-deck transit is a planning item that belongs in the concept, not a footnote discovered in week two.

The second reality is that a parking structure is never empty, and this changes what coverage means. Occupied bays are simply obstacles: the machine cleans the lanes and the free bays and moves on. A bay that is occupied tonight is free on another run, so coverage on a parking deck is built up over repeated passes rather than achieved in a single one. That is an argument for frequency over intensity, and it is also why the driving lanes deserve first priority in the route plan — they are almost always free, and they carry most of the abrasion in the building. It is worth setting that expectation with whoever signs the contract, because "the whole deck, every run" is not a thing any machine can promise on an occupied structure.

The operating window follows from occupancy and weather rather than from the usual night-cleaning habit. In an office or shopping-centre garage the classic night and weekend windows work well. In a residential garage they are exactly wrong, and mid-morning is the productive slot — which raises the noise question, and the honest answer is that no official dB figure exists for this machine, so a daytime run next to occupied flats is something to test on site rather than promise on paper. We work through window selection in our guide to night operation. Weather is the second scheduler: the water-drop sensor moves the robot to safe zones when it starts to rain, so a wet night on an open top deck is not a productive night, and the −10 °C figure is the documented lower end of the operating band rather than a statement about ice or about what road salt does to a machine over a winter.

Finally, the infrastructure that decides whether the robot actually saves labour: a mains socket and a place for the charging station on the deck it serves, a route for moving the machine if the ramp rules out autonomous transit, and somebody who empties a 35-litre container when it is full. That last one is not a detail — a sweeper that fills its hopper halfway through the run and waits is a machine that produced half a shift. Which of those tasks stay with the customer and which stay with us is exactly what our full-service model is for.

The split is clean, and it is worth stating plainly because it is the point most often oversold in this vertical. A dry sweeper removes what lies loose on the deck: grit, leaves, cigarette butts, packaging, dust. It does not remove what sticks to it. Oil films, burnt-in tyre marks and the dark abrasion layer that has bound with oil into the concrete are a wet, mechanical job, and no amount of sweeping frequency turns a sweeper into a machine that does it. Anyone offering you a sweeping robot as the answer to tyre marks is describing a machine that does not exist in this class.

That wet job is also not an autonomous job today, at least not with the scrubbers we run ourselves — for the reasons set out above, that class is released for indoor hard floors and stops at the ramp. The Adlatus CR700 does name car parks as an official application field, but it is newly onboarded with us and that claim so far stands on its datasheet alone. So the wet cleaning of a park deck stays periodic specialist work on its own cycle: ride-on scrubbers or high-pressure equipment operated by people, planned as a deep clean rather than as a recurring pass. Two site-specific questions belong in that specification and are the operator's to answer, not ours: where the wash water goes, since it carries oil and tyre abrasion and therefore sits inside the building's drainage arrangement, and what the deck coating tolerates in terms of mechanical treatment and cleaning chemistry. Both are answered by the building's own documentation and the specialist contractor. We are describing operational practice here, not giving legal or contractual advice.

There is a machine for the parts of a parking structure that people walk through, and it is deliberately not autonomous. The Pudu SH1 is an operator-guided walk-behind scrubber — never a robot, always with a person on the handle. Officially it works at 44 cm width with a 49 cm squeegee, 27 kg of brush pressure at 350 rpm, 4-litre fresh and waste water tanks, 70 minutes of run-time in standard mode and 100 in ECO, and 2.5 hours of charging. Its noise figures conflict between Pudu's own documents — 72 and 69 dB(A) on the current official product page against 76 and 71 dB(A) in the official datasheet — so plan with the higher pair until the type plate says otherwise. The decisive detail for this vertical is the floor list Pudu publishes for it: tile, terrazzo, granite, marble, epoxy finish and hardwood. Bare concrete is not on that list. The SH1's place in a parking structure is therefore the tiled or coated pedestrian zone — lift lobbies, stairwell landings, the area around the payment machines — and not the deck itself. Our machine profile covers it in full, and the scrubber-dryer guide explains which floors tolerate wet cleaning at all.

And then there is everything above ankle height, which no machine in any of these classes touches: stair treads and handrails, lift doors and call panels, signage, payment and barrier equipment, waste bins, drainage channels and gullies, and the corners a round machine cannot reach. In a parking structure that list is not small, and pretending otherwise is how a robot deployment ends up disappointing everyone. The honest framing is the one worth putting in the contract: the robot owns the recurring dry sweep of the lanes and the reachable bays, people own the rest, and the specialist wet cycle keeps its own place in the plan.

Total area is the least useful number in this vertical, and it is usually the only one on the enquiry. The list that actually decides a deployment is short and physical: the gradient of the steepest ramp; the narrowest passage the machine has to pass; the floor of the deck and whether it is bare or coated concrete; the temperature at the coldest point of the structure in January; how open the deck is, level by level, since a top deck and a basement level are different jobs in the same building; the occupancy pattern by hour and by weekday; where a mains socket sits on each level; and the split between deck area and tiled pedestrian area, because those two go to different machines. Measure those, and the machine question answers itself. Our overview of robotic sweepers compared sets the dry class side by side, and which cleaning robot fits which operation walks the whole decision from the floor upwards.

On cost, we publish factors rather than figures, and in a parking structure the factors are unusually easy to name. Area per deck multiplied by the number of decks sets the raw scale. How much of that area is actually reachable — occupancy again — decides how much of it converts into cleaned square metres. Open versus enclosed drives frequency, because wind-blown debris and a winter of grit are recurring loads that an underground level does not have. The number of ramps and whether the robot can transit them alone determines how much human handling each run costs. Then come the operator's target frequency, the share of the work that stays specialist wet cleaning, and the on-site logistics of emptying and charging. Our article on what cleaning robots cost explains how those factors interact in general terms.

One expectation deserves correcting before it becomes a disappointment. A robot doing the recurring dry sweep does not remove the specialist wet cycle from the budget. What it plausibly changes is how often that cycle is needed, because abrasion and grit that get swept out weekly have less opportunity to bind into the surface than abrasion left to sit for months. That effect is real in principle and site-specific in size, and we would rather measure it on your deck than assert a number for it. Which is the honest end point of this article: book a trial run on one level, measure what one machine completes in one window with your ramps, your occupancy and your soiling, and plan the rest of the structure from that number. If you want the funnel view of the vertical first, our parking garage page sets out how we deploy there.