Averaging

If errors were random and independent of each other, the average of enough readings would be a good estimate.

But motion isn't completely random and successive readings aren't independent.

So we try to exploit the non-independence by keeping track of how well we think we are doing at each moment.