Free, occupied, or unknown
Most reactive obstacle avoidance algorithms need a fast answer to one specific question, over and over: is this particular spot near the robot safe to drive through right now? The occupancy grid is the standard practical way to answer that. It's a grid overlaid on the robot's local surroundings, where each cell is marked free, occupied, or unknown based on the robot's current sensor readings. A cell is free if sensors show clear space there, occupied if something solid was detected, and unknown if no sensor has looked there yet or the view is blocked.
This grid is continuously updated as the robot moves and senses its surroundings. New sensor sweeps mark cells occupied when they detect something, clear cells back to free when a previously blocked view opens up, and shift the whole grid's relevant window as the robot travels so it always represents the space around the robot right now, not where the robot used to be.