State estimation (Kalman filter): The idea
In plain words
The idea explained without jargon
A moving thing, such as a spacecraft or a car, has a true position and speed that nobody sees directly. Instruments give noisy readings, and a model of how it moves gives a prediction that slowly goes wrong. State estimation blends the two. At each step the filter predicts where the thing should be, compares that with the new reading, and corrects the prediction by a share of the difference. The share depends on trust: a precise sensor pulls the answer towards itself, a noisy one pulls less. The filter also keeps track of how unsure it is, so it needs only its latest estimate, not the whole history. When the model is linear and the noise is bell-shaped, this blend gives the smallest possible average squared error [5].
as of 2026-11-12
Key equation
The equation that sums it up, and what each symbol means
- Equation
- x_new = x_pred + K (z - H x_pred), with K = P H^T (H P H^T + R)^-1
- Symbols
- {"x_new":"the improved estimate of the state (position, speed and so on)","x_pred":"the estimate predicted from the model before the new reading","z":"the new measurement","H":"the matrix that turns a state into what the sensor would read","K":"the gain, the share of the surprise that is believed","P":"the uncertainty (covariance) of the prediction","R":"the noise (covariance) of the sensor"}
- Note
- With one number to track and a sensor that reads it directly (H = 1), K is P / (P + R): a confident prediction (small P) moves little, and a noisy sensor (large R) is trusted little. The result is optimal only for linear models with Gaussian (bell-curve) noise [5].
as of 2026-11-12
An analogy
A picture from everyday life that makes the idea easier to hold
Steering a boat through fog. Your speed and compass heading say where you ought to be, but the guess drifts. An occasional glimpse of a buoy gives a position, but a blurred one. A sensible skipper leans on the glimpse when it is sharp and on the dead-reckoning guess when it is not. The picture stops being true because the skipper judges by feel, while the filter uses exact weights, assumes errors are bell-shaped and the motion is simple, and can go wrong when they are not.
as of 2026-11-12
Sources
- Constrained State Estimation - A Review arXiv (N. Amor, G. Rasool and N. C. Bouaynaya)