any bug in ndt_cuda?

[PigBroA]

Dear @koide3 , and all

I tried to utilize ndt_cuda on fast_gicp package.

However, there is no update transformation when I call "align" function.
(It just follows initial value as "align" function's second argument)

I tracked the code and found "linearize" fucntion just output "0" matrices(H & b).
("linearize" fucntion is called in "step_lm" function at lsq_registration_impl.hpp)

Can I resolve the problem or do you have any solution for ndt with cuda?

Thanks.

[koide3]

Honestly, ndt_cuda might be buggy and unstable because it was written in one day without comprehensive tests. I'm considering to implement a new one in small_gicp.