landmark / feature map

A landmark or feature map is a map that records only a scattering of distinctive points a robot can reliably spot and re-find — the corner of a wall, a table leg, a pole, a bright sign — rather than drawing every surface in the room. Each of these standout points is called a landmark or feature, and the map is essentially a list of them with their positions, like a treasure map marked only with a few crosses instead of a full painting of the island. It is a deliberately sparse, skeletal sketch of the world.

A robot uses such a map by matching the landmarks it currently sees against the ones already stored. If it can spot a few known landmarks and measure how far and in which direction each one lies, it can work backward to pin down exactly where it must be standing for those measurements to make sense — the same way you can fix your spot on a hiking map from the directions to two distant peaks. The map stays small because it remembers only the memorable bits and throws away the blank stretches of wall and floor in between.

This is the opposite design choice from a dense occupancy map, which paints in every patch of free and blocked space. A feature map is light on memory and quick to search, and it shines when the world is full of clear, repeatable markers. Its weakness shows in bare or repetitive places with few distinctive points to anchor to, and in the hard job of being sure that the landmark you see now is truly the same one you logged before, not a look-alike — a puzzle called the data-association problem.