🚀 go-pugleaf

In article <1jcou3INNped@shelley.u.washington.edu> thinman@netcom.com (Technical
ly Sweet) writes:
>
>diego@minerva.st.dsi.unimi.it (Diego Montefusco) writes:
>
>>That's exactly what I thought a Kalman filter did... I said I'm
>>completely new to this kind of things, so I'm probably in error.
>
>I don't know either.  Chris Shaw or Mark Green at U. of Alberta
>might care to comment; their MR toolkit has had a Kalman filter
>tool for a long time.  How useful is it, guys?  
>
>
>-- 
>
>Lance Norskog
>thinman@netcom.com
>Data is not information is not knowledge is not wisdom.


A key part of a Kalman filter is the noise model, this is basically
a stochastic process that models the noise and signal that you are
dealing with.  With an appropriate model a Kalman filter does a very
good job of predicting over a short range (several samples), but with
an inappropriate model you are much better off with no filtering.  Our
major contribution was not the idea of using Kalman filtering (other people
had tried it before, and found that it didn't work), but a could model
to use in the filter.  With the filter on it is possible to play real-time
games (like handball) that require reflex action and good hand-eye
coordination, without the filter the game is much harder to play.  Our
handball demo is our main test of quality of interaction.  Over the years
we have needed to change the rules, as our toolkit got better, the
game got easier to play, so we needed to make it harder to keep the
interest.