The second term of Self-Driving Car Engineer Nanodegree devotes Robotics. Therefore, the first two projects we spend on learning Kalman filter (KF) and its variations. We implemented three different versions of KF suitable for SDC and I decided to write and overview which describe key differences.
The first question is why we need KF at all. Why can’t we rely on measurement we receive from our sensors? The answer is simple. We do not live in a perfect world, and we can not trust our sensors and measurements 100%. And KF gives us a way to combine measurements from different sensors (like LIDAR or RADAR) and mathematical model we built to predict out position. How does it do it? Basically, it finds a weighted sum of our measurements depending on how much we trust a particular sensor or our model.