A glass facade robot typically leaves 3 to 7 percent of the surface untouched. The misses cluster at four places: inside corners, behind vertical mullions, the top 20 cm under a cap rail, and the bottom edge above the ground seal. Plan a manual touch-up pass for those zones.
Why do corners defeat brush geometry?
A roller brush is a cylinder. A right-angle corner is the one shape a cylinder cannot reach, no matter how much suction holds the chassis in place. On a rectangular façade the robot can clean the whole field and still leave a 5 cm dark line down every internal corner. That is not a fault, it is physics. The workaround is a second tool: a handheld pole with a soft pad for the corner runs, done by one person on a gondola or from a hatch. Budget 15 to 25 minutes per corner stack on a 30 metre elevation.
| Zone | Typical miss rate | Fix | Passes needed |
|---|---|---|---|
| Open field | 0-1% | None | 1 |
| Internal corner | 60-80% | Handheld pad | 2 per year |
| Behind mullion | 15-25% | Low flow, slow speed | Monthly |
| Top cap rail | 100% (sensor margin) | Manual band | 2 per year |
| Bottom ground seal | 100% | Manual band | 2 per year |
Do mullions hide dirt or protect it?
Vertical aluminium mullions are the quiet problem. Water and dust sheet down the glass and stop against the mullion profile, forming a grey band that the brush only partly lifts because the bristles ride over the frame. On anodised frames the band is invisible from the street and obvious from inside. Rinse the mullion line at lower flow and higher passes, or accept a monthly detail pass. Curtain wall with 15 cm reveals is worse than flush glazing, and warm-tinted coatings show the residue sooner.
What is the top and bottom edge penalty?
The top 20 cm under a cap rail and the bottom 30 cm above a ground seal are the two zones robots handle worst. Sensors keep the machine away from the parapet for safety, which is correct, and that safety margin stays dirty. On a 40 metre facade the cap-rail band alone is roughly 8 m² of glass. Multiply that across four elevations and you have a visible grime ring at the very top of the building, which is exactly where clients look first. Clean those two bands by hand twice a year, and the robot cycle will not look patchy.
How do you measure coverage instead of guessing?
Stop arguing about whether the robot did a good job and start measuring. Mark a 10 m² test panel with removable tape, run one cycle, photograph it from inside and outside, and count the untouched square centimetres. Do this at a corner, mid-panel and near a mullion. You will get a real coverage figure for your building, usually between 93 and 97 percent. Anything under 90 percent means your brush pressure, speed or water flow is wrong, not that the machine is bad.
How often should you re-measure coverage?
Coverage drifts. Brushes wear, suction cups harden, and a machine that hit 96 percent in spring may sit at 92 percent by autumn on the same panel. Re-run the taped test every quarter and after any brush or cup change. Trending the number over a year tells you when to replace a consumable before it visibly fails. It also gives you evidence for the client when they ask whether the glass really got cleaned. A falling coverage figure is an early warning, and it is cheaper to act on than a complaint.
Photograph the test panel from inside and outside each time and file the images with the run log. Raking light through the glass shows streaks a street-level photo will never reveal, and the inside view is what the tenant sees.
What about patterned or fritted glass?
Fritted glass, with its ceramic dot pattern, is a different animal. The dots protect the surface from visible streaking but trap dust between them, and a brush pass can leave a faint grid where the pattern stops. It is not dirty in the usual sense, but it looks uneven under certain light. Reduce brush pressure and rinse more, and accept that fritted glass never looks as uniformly clean as clear glass. Buyers who expect mirror-perfect fritted glazing are setting themselves up for a difficult handover.
If the pattern sits only at the edge as a band, treat that band as an edge zone and clean it by hand. The open field in the middle behaves like ordinary glass and a robot handles it well.
Related reading: suction cup, brush and filter maintenance guide, and the glass, slope and height selection guide.
Key Takeaways
- Expect 3 to 7 percent of a facade to stay dirty after a robot pass; corners and mullions account for most of it.
- Internal corners are geometrically unreachable by a cylindrical brush and always need a handheld touch-up.
- Sensor margins at the cap rail and ground seal leave the two most visible bands of glass untouched.
- Measure coverage on a taped 10 m² panel; below 90 percent means the settings are wrong, not the machine.
Do robot cleaning patterns change the coverage result?
Route matters. A machine that follows simple horizontal bands leaves the same corner edges untouched every cycle, and grime compounds there. A boustrophedon pass plus an outline pass around the perimeter lifts coverage by a few points at no extra water cost. Some operators run the perimeter twice and the field once, which is a cheap way to protect the visible edge. If your coverage test shows clean centres and dirty frames, the fix is a routing change, not a new machine.
Keep it simple: perimeter, then bands, then a slow final pass across the top band. Programme it once and every cycle repeats it. Routes that depend on the operator’s mood give you a different result each week, and you can never prove quality to a client that way.