Robot window cleaners work on flat, smooth glass between roughly 0° and 90°, on panes larger than about 0.5 m², with frames the drive wheels can register against. They fail on textured or antique glass, on painted frit, on glazing sealant grime, and on panes under 40 cm where the unit cannot turn. That covers maybe 80% of commercial glazing and rather less of domestic glazing.
The honest version of this answer is that the technology is not in question. The question is which glass you have. Everything below is about matching the machine to the surface, because that is where the money is lost.
Do they really clean, or just smear water around?
They clean when two conditions hold: enough solution contact time to lift dirt, and enough passes to remove it. On a vertical pane, solution runs off in under four seconds, so the brush has little working time. That is why a machine that cleans a bathroom window well can leave a vertical shopfront looking worse. The fix is a slower pass with more solution, which most consumer units do not allow you to set.
On horizontal glass the numbers are much better. Solution sits, the brush lifts, the squeegee removes. A single pass on a 15° skylight with a commercial unit returns a genuinely clean pane, and a second pass picks up the water film at the frame. This is the case the machines were engineered for, and it is also where the honest savings sit.
Which surfaces should never be given to a robot?
| Surface | Verdict | Reason |
|---|---|---|
| Flat float glass, 4-12 mm | Works | the designed case |
| Antique / drawn glass with ripple | Fails | wheels lose contact, unit stalls |
| Ceramic frit pattern | Risks damage | abrasion polishes the frit |
| Glazing sealant joints | No effect | grime is bonded, not loose |
| Panes under 0.4 m | Fails | no room to turn, unit wedges |
| Laminated with delaminated edge | Never | suction pulls the film |
| Coated low-E (outdoor face) | Usually fine | check brush hardness with the glass maker |
The low-E row deserves a note. Most low-E coatings are on the indoor face of the outer pane, so an outdoor robot never touches them. On some products the coating is on the exposed side, and a stiff brush over years will wear it. Ask the glazing supplier for the coating position before you put any brush on the roof.
What about the safety question people actually worry about?
Fall risk is real but lower than most fear. Suction failures happen mostly in three places: at the frame where the seal breaks and re-seats, on wet glass with a poor seal profile, and rarely at a very high glazing temperature. A rated tether removes the consequence, which is why every commercial unit ships with one. A domestic unit without a tether, on a window above a street, is a bad idea regardless of the suction figure.
The other safety matter is human. Placing a 20 kg unit on a glass roof at 5 m needs the same safe access as any other roof task. A robot does not remove the access hazard for the person; it removes the time they spend elevated. That difference still matters, because duration is a large part of exposure risk.
Who is the robot actually for?
Three groups. Facilities teams with repeated cycles on accessible glass who can put one operator on it. Owners of atria and canopies where rope access is expensive or aesthetically impossible during trading hours. And anyone with a genuinely awkward run of high windows where a pole will not reach and scaffolding is absurd.
Not for: period buildings with historic glass, anyone expecting the sealant lines to come up clean, and one-off jobs where renting a crew for a day is cheaper than a year of machine ownership. If you are in the second group, no product can help; the joint grime has to be cleaned by hand every four to six years.
For a worked example of what a credible pass looks like on a standard skylight, our Lingkong K3 review shows the pass count on a 25° glass bay. The skylight robot review archive covers the rest of the field, including the small units that get sold as universal solutions and are not.
How do results compare with a professional crew?
On clean, flat glass, a commercial robot gets within about 90% of a careful crew’s result, and the difference is almost entirely at the frame where a human hand works the squeegee into the corner. On glass that has gone six months without cleaning, a crew wins clearly because it pre-soaks and scrapes. Robots are maintenance tools, not restoration tools, and buying one to fix a neglected roof is the most common disappointment in the category.
The comparison that matters commercially is repeats. A crew that comes twice a year leaves glass that needs scraper work by visit two. A robot that runs every five weeks never lets deposits bond, so each cycle is easier than the last. Over three years that difference is larger than any difference in a single pass. This is why the payback argument works even when the per-pass quality is slightly lower.
What is the failure rate in practice?
Across commercial fleets, the usual pattern is one in-service fault per 300-500 m² of cleaning, most of them recoverable in minutes: a dock misalignment, a tether jam, a filter warning. Serious faults, meaning a unit down for more than a day, run at roughly one per two to three months of regular use on a single machine. That is a manageable service profile, provided you hold the three spare consumables named earlier and have a technician who answers the phone.
The faults that end programmes are not mechanical. They are organisational: nobody assigned to run it, no documented dock position, no spare brushes. A machine that sits in a store room for four months has a near-zero fault rate and a near-zero benefit. Assign ownership before purchase, in writing, with a name.
If you are weighing the decision against staying with a crew, the question is not whether robots work. It is whether you can keep one running on a schedule for three years. That is a staffing question with a technical wrapper, and it decides the outcome more reliably than any spec sheet.
Key Takeaways
- Robots work on flat glass from 0-90° with panes over about 0.5 m²; that is roughly 80% of commercial glazing.
- Vertical glass gives under four seconds of solution contact, which is why vertical results are often poor.
- Textured glass, ceramic frit, sealant joints and panes under 40 cm are the four clear failures.
- Tether every machine that works above 2 m, whatever the suction rating says.
- For one-off jobs, a crew for a day still costs less than a year of machine ownership.