Robust Kernels in Generalized ICP and Metric Comparisons

[MichalisPapadakis]

Hello
This tutorial mentions that Robust Kernels "have been only implemented for the PointToPlane ICP". However, when I tested them with Generalized ICP, I still got results, and the transformation differed from using the default kernel.

So, I have two questions:

question 1:
Have the kernels been implemented also for other types of ICP?

question 2:
Can i compare the fitness_ or inlier_rmse_ using different kernels to see what settings best fit my application, or the comparison between them does not make sense?

I am asking this, as both fitness_ and inlier_rmse_ in RegistrationResult mention "inlier correspondences". However, the way i understand the iteratively reweighted least-squares (IRLS) as described here, is that it just adjusts the weight of the outliers, without removing them for the ICP procedure.

I’d love to understand how the kernel impacts these metrics. Any insights would be appreciated!