Kalman filter (state estimation)
also: state estimator · observer · sensor fusion
An algorithm that combines a model with noisy measurements to estimate a system's state better than either alone; the basis of most sensor fusion.
Sensors are noisy, slow or indirect. A Kalman filter predicts what the state should be from a model, then corrects that prediction with each new measurement, weighting each according to how much it is trusted. The result is a smoother, more accurate estimate, including quantities you can't measure directly: velocity from position, a battery's state of charge from voltage and current, orientation from accelerometer and gyroscope data.
It underpins sensor fusion in drones, phones, wearables, robots and battery management systems. Variants such as the extended and unscented Kalman filters handle nonlinear systems.
Where it bites: tuning the noise assumptions takes care, a poor model produces confident but wrong estimates, and the math has a real cost on a small microcontroller. Simpler complementary filters are often good enough for basic orientation tasks.
Founder takeaway
If your product needs a quantity no single sensor gives cleanly, ask whether an estimator is planned and how its accuracy will be validated.
engineer review pending