simultaneous localization and mapping (SLAM)

Simultaneous localization and mapping, almost always shortened to SLAM, is a robot solving two tangled problems at the very same time: drawing a map of a place it has never seen, while also keeping track of where it is on that very map. Imagine being dropped, blindfolded, into a pitch-dark unfamiliar house and asked both to sketch its floor plan and to mark your own position on it as you feel your way around. You have no map to locate yourself against, and no fixed position from which to draw the map — yet you must produce both, each leaning on the other.

That circular dependency is the whole difficulty, and it is also the trick that makes SLAM possible. As the robot moves, it spots features around it — a corner here, a pillar there — and notes how far and in which direction each one lies. From those bearings it can both refine where it thinks it is and pin those features onto its growing map. The two estimates prop each other up, like two people steadying themselves by leaning together: a slightly better idea of your position sharpens the map, and a slightly better map sharpens your idea of your position. The catch is that small errors creep in with every step and quietly accumulate, so the map can slowly warp.

The cure for that drift is a moment SLAM systems prize, called loop closure: when the robot revisits a place it has been before and recognizes it, that recognition snaps the accumulated error back into line and tightens the whole map at once, the way checking your work against a landmark you already trust corrects all the guesses in between. SLAM in its modern forms — using laser scanners in LiDAR SLAM, or ordinary cameras in visual SLAM — is the technology under the hood of robot vacuums, self-driving cars, drones, and the augmented-reality apps that place virtual furniture in your living room.