Introduction
A microcontroller is often used to interface with sensors and generate control actions based on information
gathered from sensors. For reliability and completeness, more than one sensor is generally used. If different
sensors are employed to provide measurement of the same physical quantity, a better estimation can often be
obtained by calculating weighted average of the individual measurements, assuming the noises they carry are not
statistically correlated.
Sometimes it is useful to have sensors that measure different physical variables and the combined information is
used for control decision making. The operation to combine information from such multi-modal sensors is called
sensory fusion.
Kalman filters can be used to derive the best estimation by combining sensory input from different sources. It has
been used most extensively in navigation since it was invented in 1960. Recently, it has been used in high-end
GPS navigation systems to improve performance.
The concept of Kalman filtering can be illustrated in Figure 1. Suppose a human is traveling in one dimension. He
has a positioning device that gives him estimations of where he is at time t1, t2, and t3, with some variance. For
example, if the measurement says the person is at position t2, the real position could be anywhere around t2 with
a certain distribution function as shown in the figure.