Implementing Isaac ROS' CuVSLAM visual odometry strategy as a 3rd party strategy in RTAB-Map.

cuvslam headerfile:
https://github.com/NVIDIA-ISAAC-ROS/isaac_ros_nitros/blob/main/isaac_ros_nitros/lib/cuvslam/include/cuvslam.h

To run with CuVSLAM

Set the following environment variable so the system versions of libpng16, libjpeg, and libjsoncpp are loaded before cuVSLAM’s bundled copies. This avoids runtime crashes caused by conflicting dynamic libraries:

VAL="$(ldconfig -p | awk '/lib(png16|jpeg|jsoncpp)\.so/{print $NF}' | sort -u | tr '\n' ' ' | sed 's/[[:space:]]*$//')"
export LD_PRELOAD="$VAL"

cuVSLAM prerequisites (Ubuntu 22.04, x86_64)

1) NVIDIA Driver + CUDA 12.6 (toolkit)
https://developer.nvidia.com/cuda-12-6-0-download-archive?target_os=Linux&target_arch=x86_64&Distribution=Ubuntu&target_version=22.04&target_type=deb_network

2) NVIDIA VPI
https://docs.nvidia.com/vpi/3.0/installation.html

3) Isaac ROS Visual SLAM (NITROS/cuVSLAM)

sudo apt-get update
sudo apt-get install -y ros-humble-isaac-ros-visual-slam