Combining imperfect information
Every sensor lies a little. A camera can be blinded by sun glare. LiDAR can misread a puddle as open space because the laser reflects off it strangely. Wheel encoders that estimate distance traveled slowly drift as tires slip. Sensor fusion is the practice of combining readings from multiple sensors so that the weaknesses of one are covered by the strengths of another, producing an estimate that's more reliable than any single input.
A self-driving car is the clearest example: it typically carries cameras for recognizing lane markings and traffic signs, radar for tracking the speed of other vehicles even in rain, and often LiDAR for precise 3D shape detection. Each sensor votes on what's happening around the car, and the fusion software weighs those votes based on which sensor is most trustworthy in the current conditions.