50.1.1 Aim and main principle of Kalman filters

Source: rlabbe

Aim

Aim of the Kalman/Bayesian filters: to accumulate (or to somehow blend)

• our noisy and limited knowledge (of system behaviour)
• noisy and limited sensor readings

and with these, make the best possible prediction (estimate) of the system state.

Main principles:

• use past information to make predictions for the future
• never throw away information
• predict/propagation step: calculate prediction based on process model and using previous state data (previous estimate)
• update step: calculate the estimates based on prediction and measurement

Prediction step a.k.a. evolution One iteration of prediction + update is called an epoch