A Non-Mathematical Introduction to Kalman Filters for Programmers

> Indeed, if you had a sensor that you could guarantee never goes offline and measured what you wanted to know with an arbitrary degree of accuracy, there’d be no need for Kalman Filters at all.

Even then, there are cases where using a Kalman filter can be an improvement. Sensors have latency between when they measure things and when you receive the measurement (for example, a pressure sensor might average measurements over a few ms too report more stable readings). If that delay is large enough, you may be better off using a model to predict the sensor data than to wait for 100% reliable but (somewhat) outdated information.