A window cleaning robot works by creating vacuum between its base seal and the glass, then walking on caterpillar tracks while spraying water ahead of a wiping pad. Edge sensors stop it at the frame. A battery holds vacuum if power fails. It cleans 300-400 mm per pass at 6-12 m²/h.
“Window cleaning robot how does it work” is the query that should come before the purchase, and usually comes after. Fine. Here is the mechanical truth, including the parts sellers skip.
What holds the robot on the glass?
A centrifugal or piston pump pulls air out of the cavity under the chassis. The pressure difference between inside and outside the seal is what presses the machine against the pane. Ratings between 2,500 and 4,000 Pa are typical. On a flat vertical window, that is more than enough: the holding force can exceed several times the unit’s 1.5-2.5 kg weight.
Movement comes from two silicone tracks. They need grip on a wet surface, which is a harder problem than holding still. Some units also add a secondary vacuum cup that lifts and reseats to break surface tension. When people report a robot “losing grip”, it is almost always the track slipping, not the vacuum pump failing.
How does it know where the window edge is?
Four or more optical or mechanical edge sensors sit near the corners of the chassis. When one detects the frame, the controller reverses that motor and starts a new line. Simple, effective, and completely dependent on the frame being visible.
This is why frameless glass, mirrored walls, very dark tinted panes, and heavily patterned film confuse some machines. The sensor reads the change in reflectivity and finds nothing. On a glass roof with structural bars, the reverse happens: every bar reads as an edge and the robot treadmills in one bay. If you have a bar-covered skylight, check how a model handles bars before you buy; our Lingfeng S1 review goes into the crossing behaviour specifically.
Where does the dirty water go?
Two-tank systems spray clean water ahead and vacuum dirty water back behind the pad. Single-tank or no-tank units just wipe with a damp microfibre pad and leave the runoff to gravity. The second is cheaper and, on an indoor pane above a marble floor, a genuine hazard.
Water rate matters too. A common figure is 0.2-0.4 L per 10 m². Run the pad dry and you are not cleaning, you are polishing dust into the glass. Run it too wet at 30 degrees of tilt and you are feeding the exact film that makes tracks slip. The first pass should be wetter than the second; most units let you set that.
Why do some robots fall off?
Three causes, and none of them are mysterious.
- Loss of mains power without a working hold battery. The vacuum stops, the robot detaches, the tether catches it or does not.
- Seal breaking on an obstacle. A raised glazing bar, a window catch, a fly-screen rail. The seal ring lifts at one corner and holding pressure drops below the weight.
- Incline beyond the design limit. The stated limit is almost always for dry glass. Wet glass at the same angle is a different, worse number.
Every one of these is testable before purchase. Ask for the incline rating on wet glass, the hold time after power loss, and the maximum step height the chassis can cross. If the answers are vague, the machine is a flat-window tool and should be sold as one.
Do window cleaning robots really work, day to day?
Yes, within a narrow band. On flat vertical glass, indoors or out, with a clean frame and a power socket nearby, a robot does a competent job and does it weekly without a phone call. That is worth real money to a facilities manager.
Outside that band it gets complicated. Glass roofs, atriums, curtain wall above 30 m, anything with a step or a bar. The honest split is that a suction robot is a maintenance machine for accessible glass, and a facade or skylight system is machinery for the rest. Different category, different budget.
If you want the fuller picture on which models hold up in testing, the skylight robot review archive covers grip, runtime and edge behaviour on real buildings rather than showroom panes.
How does the cleaning pattern affect the result?
Most units run one of two paths: a zigzag down the pane, or a spiral that starts at a corner and works inward. Zigzag is faster and leaves overlapping edges where the pad reloads with dirt. Spiral keeps the pad on clean glass longer and gives a more even finish, at the cost of time.
The bigger variable is pass direction. A pad dragged down a pane leaves a thin water line at each stop point; a pad dragged sideways leaves streaks that follow the sun. On north-facing glass in Europe it barely matters. On a west-facing curtain wall in summer, sideways streaking is visible from the street at 5 pm. If the finish is being photographed, run the second pass downward.
Two passes, not three. A third pass on dry glass builds static and drags dust back onto the surface. If the glass still looks dull after two, the problem is the water, not the count.
What about noise, and does it matter?
Pump noise runs around 60-70 dB at a metre, similar to a conversation in a busy room. That is fine on an empty facade and annoying inside an office where someone is on a call. Interior partition cleaning is usually scheduled outside working hours for this reason alone, not for the dirt.
Key Takeaways
- Vacuum plus tracks hold and move the robot; 2,500-4,000 Pa is the usual holding range.
- Edge sensors rely on visible frames; frameless and bar-covered glass cause navigation errors.
- Most “robot fell off” reports trace to power loss, seal break on an obstacle, or excess incline.
- Efficiency runs 6-12 m²/h, so a 200 m² facade is a fleet job, not a single-unit job.