2
$\begingroup$

I have read in SLAM papers talk about marginalization. I am confused about how this process works, why you do it and how it differs from global smoothing.

In particular if you're interested in Windowed Smoothing I don't understand how marginalization is different than full bundle adjustment? If marginalization carries the full information from the prior states, then why do global bundle adjustment?

$\endgroup$
2
  • $\begingroup$ From my understanding, you remove older nodes, but introduce additional constraints in order to retain most of the information the nodes contained. Briefly speaking, this will make optimization faster in exchange for a little accuracy. Here's a more detailled answer robotics.stackexchange.com/a/15482/33402 $\endgroup$
    – Philip Z.
    Commented Jun 26, 2023 at 8:00
  • $\begingroup$ Hi @PhilipZ. - I rejected your suggested edit as it seems to alter the meaning. I believe that the OP means If marginalization carries the full information from the prior states, then why [bother] do[ing] global bundle adjustment [as well]? and not the opposite which your edit would have implied. However, we both could be wrong - only the OP would actually know..! :-) $\endgroup$ Commented Jun 26, 2023 at 9:32

0

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Browse other questions tagged or ask your own question.