Kalman Filter

You need the Kalman filter to track the position of something moving—like a car, a satellite, or even a stock price—when you only have imperfect, noisy measurements. Without it, your tracking would jitter wildly with every new data point. It is the standard method used in GPS navigation, autonomous driving, and robotics to smooth out sensor errors and provide a clear, reliable path. Learning this gives you the practical tool to turn messy real-world data into stable, usable estimates.

Imagine you are trying to locate your friend in a crowded market. Your friend sends a text saying they are near the chai stall, but their GPS might be off by a few meters. You also have a rough guess from where you last saw them standing. The Kalman filter works by combining these two pieces of information: it takes your prediction (where you think your friend is) and your measurement (the GPS text), and blends them together. It weighs each based on how confident you are in that specific source. If the GPS is known to be very accurate, you trust the text more. If you know your friend moves erratically, you might trust your prediction of their path more than the shaky GPS reading. The result is a single, smoothed estimate of where your friend is now. This process happens in a repeating cycle: you use this new best estimate to predict where they will be in the next moment, then combine that with the next noisy measurement. Because it uses its own previous output as the starting point for the next step, this cycle is called recursive.

Think of tracking a bird flying across a field with a companion who is very good at predicting motion. You have two inputs: your intuition about where the bird should be based on its last known speed and direction (the prediction), and occasional photos from a camera that captures the bird’s location but is sometimes blurry or slow (the measurement). Your companion doesn't just trust the photo or just trust the prediction. Instead, they act as a weighter. If the camera is sharp, they lean heavily on the photo. If the wind is gusting wildly and making the bird's path hard to predict, they lean more on the physics of the bird’s flight. They produce a single, smooth estimate of the bird's location. The analogy breaks down if the bird suddenly performs a maneuver that defies normal physics, because the filter assumes movement follows predictable patterns.

A Kalman filter is a repeated algorithm that estimates the current condition of a moving system by constantly blending a predicted position with a noisy measurement, using confidence levels to weight each source, resulting in the most accurate possible estimate for linear systems with random noise.

This concept sits within Numerical Methods, which covers how to compute mathematical answers quickly and accurately using limited computational resources. It directly relies on Probability, specifically the treatment of errors as random variations around an average (known as Gaussian distribution), and it connects to Linear Algebra, which provides the tools for handling multiple variables simultaneously through vectors and matrices.