Mapping and SLAM

What you’ll learn

Maps robots use: occupancy grids
Localization: where am I on a map?
SLAM: building the map and localizing at the same time

Occupancy grids

The world is split into small squares (for example 5 × 5 cm). Each cell holds the probability that it’s occupied: free, wall, or unknown. Distance sensors fill it in: cells along a ray are free, the cell where the ray hits is occupied.

Localization

With a map, the robot compares what its sensors see now with what they should see from each possible position. A popular method keeps many guesses (“particles”) and keeps the ones that match best: Monte Carlo localization.

SLAM

Simultaneous Localization and Mapping solves both at once, a chicken-and-egg problem: you need a map to know where you are, and you need to know where you are to build a map.

Odometry (encoders, IMU) gives a first guess of movement, which drifts.
Scan matching lines up new sensor data with the map to correct it.
Loop closure: when the robot recognises a place it’s seen before, it corrects all the drift around the loop.

Sensors for SLAM

2D lidar (a spinning laser): the classic choice for indoor robots and robot vacuums.
Cameras (visual SLAM): cheaper, harder.
Depth cameras for 3D maps.

In ROS 2, packages such as slam_toolbox build maps from lidar data, and Nav2 plans routes on them.

Example: filling a grid from one ray

The robot is at cell (10, 10) facing east, and the distance sensor reads 40 cm with 10 cm cells:

Cells (11,10), (12,10), (13,10) → free
Cell (14,10) → occupied
Cells beyond → still unknown

Example: why loop closure matters

A robot drives round a square corridor. Odometry says it ended 30 cm away from where it started, but its lidar recognises the starting doorway. SLAM shifts the whole loop’s path to close that 30 cm gap, and the map’s walls line up.