Glossary · Control theory and process control
Kalman filter
Also known as: Linear quadratic estimator
German: Kalman-Filter
In control engineering and signal processing, a Kalman filter is a recursive algorithm that estimates the state of a dynamic system from noisy measurements and a model, weighting prediction and measurement by their assumed uncertainties.
- Control theory
In one sentence
A Kalman filter recursively estimates system states from a model and noisy measurements, weighting each by its uncertainty.
Example
A mobile robot fuses wheel odometry and a laser-based position measurement in a Kalman filter to estimate its position more accurately than either source alone.
How it applies
- Engineering: The classic Kalman filter is optimal for linear systems with Gaussian noise; extended and unscented variants handle nonlinear systems. It needs a State-space model and noise covariances, which are often tuned empirically.
- Operation: Estimated values are not measurements. If a sensor fails, the filter may keep producing plausible-looking estimates for a while, so fault detection and Plausibility check functions remain necessary.
- Documentation: Label estimated signals as estimates on HMIs and in data exports, and document which measurements feed the filter. This matters for Sensor data integrity and for anyone analyzing historical data.
Kalman filter vs. simple filtering
A low-pass filter smooths one signal without a model. A Kalman filter uses a model of the system and can estimate states that aren't measured at all, such as velocity from position measurements.