SLAM (simultaneous localization and mapping)
Also known as: simultaneous localization and mapping, simultaneous localisation and mapping
SLAM, short for simultaneous localization and mapping, is the technique that lets a robot, drone, headset, or phone build a map of a place it has never seen while working out where it is inside that map at the same time.
It's what lets a robot vacuum clean in neat rows and a mixed reality headset keep virtual objects anchored to your table. While SLAM is now built into everyday devices, it can still get lost in featureless, changing, or poorly lit spaces.
SLAM (simultaneous localization and mapping), explained at three levels
At home
Imagine walking into a dark house with a flashlight, sketching a map as you go while marking where you stand on it. Each time you spot a doorway you've seen before, you correct your sketch. SLAM lets robots and devices do that: they build a map of an unknown place from camera or laser readings while working out their own position within that map at the same time.
At work
SLAM is a family of algorithms that lets a robot or device map an unfamiliar space and track its own location within it, using cameras, lidar, and motion sensors. It works where GPS can't, such as indoors, underground, or in warehouses. For businesses, it means robots, drones, and scanners can work without pre-built maps or installed markers, cutting setup time.
At home
SLAM runs a loop. The device picks out distinctive features, such as corners and edges, in camera images or laser scans. As it moves, it measures how those features shift to estimate its own motion, helped by wheel sensors or an accelerometer and gyroscope. It adds the features to its map. Small errors build up over time, so when it recognizes a place it has visited before, called loop closure, it pulls the whole map back into line.
At work
SLAM systems differ by sensor. Lidar SLAM, used in warehouse robots and survey scanners, measures distances directly and works in the dark. Visual SLAM, such as ORB-SLAM, uses cheap cameras but needs texture and light. Visual-inertial SLAM, used in phones and headsets through ARKit and ARCore, fuses cameras with motion sensors for smooth tracking. Most commercial systems split into a fast front end that tracks motion and a back end that optimizes the map.
At home
SLAM depends on recognizing landmarks, so it struggles where they are missing or misleading: long blank hallways, glass walls, dim rooms, or spaces where furniture moves. Moved objects can convince the robot it is somewhere else. While modern systems handle this better, in my opinion that's why your robot vacuum sometimes acts lost after you rearrange the living room, and why remapping is worth doing.
At work
Modern SLAM treats the problem as a factor graph: each pose and landmark is a node, each measurement a constraint, and an optimizer such as g2o or GTSAM finds the most consistent map. Learned features and neural map representations are growing. Production issues include compute limits on embedded chips, maps going stale in changing sites, sharing maps across robot fleets, and privacy, since a SLAM map of a home or plant is sensitive data.
Real-world examples
At home
Fatima's robot vacuum mapped her apartment on its first run. Now it cleans room by room and she can draw no-go zones around her pet's water bowl in the app.
At work
Carlos is a surveyor in Alberta. He uses a handheld lidar scanner with SLAM to map the inside of an old industrial building in minutes, without GPS signal.
How it compares
| feature | SLAM (simultaneous localization and mapping) | GPS navigation |
|---|---|---|
| Needs existing map | No | Uses existing maps |
| Works indoors | Yes | Poorly |
| Sensors | Cameras, lidar, motion sensors | Satellite receiver |
| Drift | Builds up; corrected by loop closure | Minimal outdoors |
SLAM works where GPS can't, but it must constantly correct itself.
For analysts
For robots in changing workplaces, I'd test SLAM performance after layout changes and in low light. Maps that worked during setup can fail once shelves and pallets move.
History & origin
Researchers including Randall Smith, Matthew Self, and Peter Cheeseman laid the statistical foundations for estimating robot position and map uncertainty together in the mid-1980s. The acronym SLAM came into use in the mid-1990s, with Hugh Durrant-Whyte and John Leonard among the key figures. Early systems used laser rangefinders and sonar. Visual SLAM using ordinary cameras matured in the 2000s, with systems such as MonoSLAM in 2007 and ORB-SLAM in 2015. iRobot's Roomba 980 brought camera-based navigation to home robots in 2015, and Apple's ARKit and Google's ARCore put visual-inertial tracking on phones in 2017. In my view, SLAM is one of the great unsung algorithms, quietly running in devices most people never think about.