Robot vacuum cleaners typically rely on algorithms for mapping and localization. They scan the environment using LiDAR or cameras to create a map of the room. Optimal cleaning routes are then calculated using these algorithms. Thanks to their sensors, they can avoid obstacles and adjust their cleaning based on the requirements of different surfaces. So, what are the situations where these systems fall short?
How do robot vacuum cleaners work?
👁️ 4 views💬 1 replies❤️ 0 likes
1 Replies
I encountered one of the weakest links in the algorithm when I saw the point cloud data on the screen while a LiDAR-based robot vacuum was mapping a 20 m² house: transparent objects. Normally, the laser beam passes through glass to detect the back wall, but in the vacuum's camera mode, it sometimes records the cat moving inside the glass as a "gap." Result: the vacuum suddenly stops in front of the glass and signals "surface changed."
In a benchmark test by Stanford, I saw that LiDAR-based systems have an error rate of around 12% with transparent objects; in camera modes, this classification error can rise to 25%.
Another issue is depth changes. When the vacuum's dustbin starts to fill up, despite the increase in motor noise, the algorithm has to adjust the "fullness" threshold due to the reduced mean free path. This calculation problem is especially noticeable in spiral mode; while the chance of hitting an elevator shaft is 3-4% at that moment, the system can't prevent "drift" and collisions with walls. The algorithm for getting rid of cable debris still gets confused with pet tails despite firmware updates—this is, I think, the most complained-about error mode by pet owners.
Finally, IR sensors used for tactile obstacle detection fail to capture enough reflection on completely black surfaces (like light-absorbing carpets), leading them to detect "no vertical obstacle," causing the robot to constantly slide off the edge of the carpet. When I experienced a sliding rate of up to 45%, I confirmed that even a perfect fan motor can't compensate for these "blind spots."