The biggest letdown

Just another quote to come out of one of our meetings:

"So I took a look at grouping some of the LIDAR data. I implemented a dynamic programming approach to fit lines to all subgroups of data, with penalties set on R^2 values. I also penalized deviations in the groupings, to favor straighter lines over corners and outliers."

Sounds great, right? Then he made one little addendum:

"It didn't work at all."

Not that I really expected surface identification to work right off the bat (heck, it may never work), but it sounded intelligent until the last part. Did I mention how hard this problem was?

***

Along a separate line of inquiry, my GPS truth model simulations are revealing an inherent bias in the GPS navigation solution. It's below the noise floor of the pseudoranges, but it appears in more precise positioning techniques. One way to remove it is with WAAS or other differential correction services, which broadcast precise corrections to raw GPS signals. Using those services is a very tempting cop out, until one realizes that they probably won't be available in the city. I don't have an immediate solution (stuff is in the works)... Until then, the search for the holy grail* continues.

*holy grail = precise positioning without differential corrections