Guide
Why glass and mirrors confuse robots.
A glass door, a mirror, a black rug: things people hardly notice can be invisible or misleading to a robot. The reason is the light most robots measure with, and the fix that research keeps coming back to is a second sense.
In short. Many robots measure distance by timing infrared light. A smooth pane of glass lets most of it through or reflects it away like a mirror, so the robot sees the room behind the glass, or nothing; a mirror shows it a room that is not there; a very dark floor can return so little light that it looks like a drop. Research robots fix this by adding a second sense, usually ultrasonic sound, and by remembering what they found. The catch is that the fixes work best head-on and in clear cases.
Why light-based sensors struggle
Laser scanners and time-of-flight sensors send out infrared light and time how long it takes to come back. Glass breaks that in two ways. Clear glass lets most of the light through, so the robot measures the wall behind it. And because a pane is smooth, the light it does reflect leaves at a mirror angle, away from the sensor, so the pane itself shows up as empty space. Only when the sensor faces the glass almost square-on does a small, bright return come straight back. A mirror goes further and shows the robot a whole room that does not exist. A shiny dark surface may send back too little light to measure at all.
Dark floors and the stairs problem
Robot vacuums look down as well as ahead: infrared “cliff” sensors under the front stop them at the top of a staircase. A very dark rug or tile reflects little infrared light, so to those sensors it can look like a drop, and some robots stop at its edge or will not cross it. Switching the sensors off would switch off the stair protection too, which is why the usual answer is a no-go zone in the app.
What research robots do
- A ground robot, 2021. Researchers at Tecnológico de Monterrey gave a small robot a 360° laser scanner, a stereo camera and two ultrasonic sensors. To the laser alone, a glass pane was transparent and the map was badly wrong, by more than two metres. A small neural network trained to correct the laser’s readings with the ultrasonic and camera data brought the error to about three centimetres against their reference map. The camera on its own did not detect the glass, and the ultrasonic sensors cannot tell direction within their wide cone.
- A 280 g drone, 2025. In a 2025 preprint, four robotics researchers fitted a sub-300 g quadrotor with a time-of-flight camera and a single ultrasonic range sensor. The sound reading cuts away what the camera sees through the glass; the method then looks for the small bright speck a pane returns when seen nearly head-on, checks that it repeats over several frames, and fills in the pane on the map so the planner treats it as solid. It ran in real time on the drone using a fraction of one processor core, where two deep-learning detectors it was compared with could not run on board. The authors say it is most reliable head-on and with a clear background, and that stickers or objects close to the glass can hide it.
What helps at home
- A second sense. Ultrasonic sensors measure with sound. When light says “far” and sound says “near”, there is probably glass or a mirror.
- A map that remembers. Once found, a glass door stays on the map, even from angles where the sensors cannot see it.
- Caution near what cannot be seen: slow down and keep a margin, larger when the robot is less sure where it is.
- Your help. Marking a glass door, a mirror wall or a dark rug as a no-go zone in the app removes the guesswork.
In Alkyon Concept
Alkyon is designed with the same pairing as the 2025 drone: light, in time-of-flight sensors, and sound, in ultrasonic sensors facing forward and down. When they disagree, it treats the spot as glass or a mirror, keeps it on its map, stops, and looks around before it goes on. So far this is tested in simulation only. How the rest of its navigation works: how indoor drones find their way; what else goes wrong indoors: flying a drone indoors.
Sources, read 4 November 2026: Barreto-Cubero and colleagues, sensor data fusion for a mobile robot using neural networks, Sensors, volume 22, published online 31 December 2021; Hopkins, Murali, Kumar and Taylor, real-time glass detection and reprojection using sensor fusion onboard aerial robots, arXiv preprint, October 2025.
Questions
Why can’t robots see glass?
Most measure distance with infrared light, and a smooth pane lets it through or bounces it away like a mirror, so the robot sees the room behind the glass or nothing at all. Only when the sensor faces the pane almost square-on does a little light come straight back.
Why does my robot vacuum avoid dark rugs?
Robot vacuums point infrared sensors at the floor to stop at stairs. A very dark rug sends back little of that light and can look like a drop, so some robots will not cross it. A no-go zone is the usual answer; covering the sensors would also switch off the protection at the stairs.
Can a robot find glass with sound?
Research robots do. In a 2021 study a small ground robot combined its laser scanner with two ultrasonic sensors to put glass on its map, and in a 2025 study a 280 g drone used one ultrasonic range reading with a time-of-flight camera to find glass panes in real time.