RTABMAP Acceleration on Jetson Orin or AGX

[Gi_T]

Hey Mathieu,

I was wondering how RTABMAP computational performance can be optimized for a Jetson Orin NX or Jetson AGX. I have seen some issues and PR request:
- https://github.com/introlab/rtabmap/issues/561
- https://github.com/introlab/rtabmap/pull/1330

but I'm not sure what all the libraries are that can be accelerated with GPU and what effect one can expect for different applications like ICP odometry or visual odometry.

I would appreciate any guidance in help of getting the most out of RTABMAP for embedded platforms. Also I think it be great to create an overview about building RTABMAP with GPU support (or a Docker container), parameters for enabling acceleration and possibly even performance difference for a few given systems but of course I'm aware that this will entail a lot of work.

Thanks in advance.

Cheers,
Gilbert