sensor fusion
Sensor fusion is the art of combining several imperfect sensors into one estimate that is better than any of them alone. Every sensor has a weakness: a camera sees rich detail but is blinded by darkness; a GPS knows roughly where you are but jitters by several meters; an inertial sensor reacts instantly to motion but slowly drifts off course. Fusion is like a jury weighing several flawed witnesses — no single account is fully trusted, but pooled together they converge on a story far closer to the truth than any witness could give on their own.
The clever part is that fusion does not just average the sensors; it weighs each one by how much it can be trusted at that moment, and lets them cover each other's blind spots. The fast inertial sensor fills the gaps between slow GPS updates, while the GPS keeps pulling the drifting inertial sensor back onto course — each one's strength patches the other's weakness. A self-driving car blends camera, radar, LiDAR, and wheel sensors this way so that fog blinding one of them does not blind the whole car. Your phone fuses its accelerometer, gyroscope, and magnetometer just to know which way is up and which way you are facing.
Doing this well is really a question of estimation under uncertainty, and the mathematical machinery that does it — the Kalman filter, particle filters, and their relatives — lives in the world of state estimation. So think of sensor fusion as the everyday name for the goal (turn many noisy senses into one clean belief about the world), and state estimation as the deeper toolbox of methods that actually pulls it off.
Walking through a tunnel, your phone's map keeps showing you moving even though the GPS signal is gone. The phone has quietly switched to fusing its motion sensors, dead-reckoning your path until the satellites come back into view.
When GPS drops out, fusion leans on the inertial sensors to keep the position estimate alive.
A rough rule of thumb: fusion shines when the sensors fail in different ways. Combining two sensors with the exact same blind spot adds little, but pairing a drifting-but-fast sensor with a steady-but-slow one gives you the best of both.