How we earn: we plan to earn commissions from some links. Today we hold no affiliate approvals, so every button goes straight to the seller. The System Builder ranks by fit and 3-year cost — never by commission. Editorial examples may feature a merchant we partner with, but only when it genuinely fits. How we make money

How robot vacuums find their way

Every robot vacuum combines two jobs: knowing where it is, and not hitting things. The sensors that do each job decide whether it cleans your whole floor in neat rows, gets lost in a dark bedroom, or eats a phone cable. Here's how each system works and where each one fails.

The short answer

A spinning laser (LiDAR) on top is common on mid-priced and premium robots. It maps in the dark, but it can't see low objects on the floor.
A camera that maps from what it sees (vSLAM) needs light. iRobot says dark rooms can cause navigation and coverage problems.
A front camera with object recognition is what lets a robot recognize and steer around socks, cables and pet waste. It adds a camera to your home.
Cliff sensors are separate infrared sensors underneath. They stop falls down stairs, and they can mistake a black rug for a drop.
No system is perfect, and makers say so: in its own navigation explainer, Roborock writes, There isn't yet a perfect system.

Where each sensor looks. The laser on top maps the room but looks over low objects; the front sensors handle obstacles; separate infrared sensors underneath watch for drops. Simplified drawing, not to scale.

Job one: knowing where it is

Bump-and-turn (no map)

The oldest and cheapest approach doesn't map at all. The robot drives until its bumper hits something, turns, and repeats, sometimes with simple patterns like spirals and wall-following. Ecovacs describes budget models as working on the bump-and-run principle. Coverage is random, so it cleans the same patches repeatedly and misses others, and it can't clean one chosen room.

LiDAR, also sold as LDS

The round turret on top of most mid-priced robots is a laser distance sensor. It spins 360°, fires laser light at the walls and furniture, and measures the reflections to build a floor plan. The robot then works out where it is on that map while it moves, a technique called SLAM. Makers sometimes call the turret LiDAR and sometimes LDS; for robot vacuums they mean the same kind of sensor.

What it does well: it builds an accurate map quickly, cleans in efficient rows, and works with the lights off, because it brings its own light. What it misses:

Low objects. The laser sits on top of the robot, so it looks over things lying on the floor. Roborock says the top-mounted sensor can't sense small objects at floor level as effectively.
Black and see-through surfaces. Dreame notes that transparent and black surfaces are hard for a laser to read. Slamtec, which makes LiDAR modules, lists glass walls, mirrors and polished surfaces among the hardest settings, because the light reflects away instead of back.

Camera mapping (vSLAM)

Some robots, including many Roombas, map with a camera instead of a laser. The camera picks out landmarks such as corners and furniture edges and tracks them as the robot moves. It needs light to see them. iRobot's own support pages say dark rooms can make navigation and coverage worse, and recommend at least daylight or dim lighting. If you want to run the robot at night with the lights off, a LiDAR model is the safer choice.

Job two: not hitting things

Front camera with object recognition

A forward-facing color camera, paired with software trained to recognize common objects, is what lets a robot tell a sock from a shadow and steer around it. This is the feature to pay for if your floors have cables, toys or pets. The trade-off is privacy: there's a camera on the robot, and some apps let you view its feed. Our pet-mess guide covers what camera avoidance can and can't promise.

Structured light and 3D sensors

Many robots project a pattern of light, often lines or dots, in front of them and measure how it bends over objects. Roborock compares it to a ruler measuring distance to obstacles. It's good at the shape and height of things ahead and works without a camera image, but very flat objects still slip through.

Bumpers and wall sensors

Nearly every robot has a physical bumper and side sensors for edges. They're the last line of defense, not a strategy.

Cliff sensors and the black-rug problem

Underneath the robot are small downward-facing infrared sensors that watch for drops such as stairs. They're separate from the LiDAR or camera, and they work by bouncing infrared light off the floor. Very dark surfaces absorb that light, so the robot can read a black rug as a cliff and refuse to drive onto it. iRobot says its cliff sensors can sometimes see dark surfaces as drop-offs, and eufy says its RoboVacs aren't suitable for black flooring or dark carpet for the same reason.

If your home has dark rugs or dark tile, check the maker's support pages for that model before you buy. Dust on the sensors causes similar errors, so wipe them during routine cleaning.

What every system still misses

Very flat objects are the common weakness. One Anker/eufy support page sets a minimum obstacle height of 15 mm on hard floors and 25 mm on carpet for its avoidance to work, and tells owners to put away data cables and socks lying flat. Eufy's own guide to obstacle avoidance lists loose socks, shoelaces, transparent or highly reflective objects, and dark mat edges as hard cases. A quick pick-up before each run still matters, whatever the box promises.

System	What it's for	Works in the dark?	Main weakness
Bump-and-turn	Cheap random coverage	Yes	No map, misses patches, no room selection
LiDAR / LDS	Mapping and position	Yes	Can't see low objects; mirrors, glass and black surfaces
Camera mapping (vSLAM)	Mapping and position	Poorly	Needs light
Front camera + recognition	Recognizing obstacles	Varies by model	A camera in your home; flat objects
Structured light / 3D	Measuring obstacles ahead	Yes	Very flat objects
Cliff sensors (infrared)	Stopping falls	Yes	Dark rugs read as drops; dust

What this means when you buy

Cluttered floors or pets: pay for a front camera with object recognition. See the camera models in our floor-cleaning guide.
No camera in the house: a LiDAR model with structured light is the usual compromise. Our guide has a no-camera list.
Dark rugs: check that model's cliff-sensor guidance first.
Want it matched for you? The System Builder asks about floors, pets and cameras and returns a complete system with a 3-year cost.

Related: what docks do · robot glossary.

Sources

All sources read October 12, 2026. These are the makers' own descriptions of their technology; we haven't run our own sensor tests.

Last checked October 12, 2026. See what changed.