Experiments with Kalman Filters for Navigating with Sonar

I had the idea to aid long-term navigation of Xavier by registering 
local sonar data in accumulated evidence grids. I'll show you how 
that went (not so good). To fix these problems, I resorted to using 
an extended Kalman filter to integrate the local sonar data with
a state estimate, rather than with the evidence grid. The results were
very clean evidence grids, as compared to before. I'll talk about the
details of that, if people are interested. I'll also show how I used
the Hough transform to find straight wall segments, in order to
optimally decide which sensor data to integrate, and which to discard.
I'll end by solliciting a discussion about the value and potential for
the techniques I outlined.