I am testing this package which is ORB_SLAM 2 but revised and cleaned:

LINK

From the launch file rs_camera.launch I can see that the author simply creates TF between cameralink and map. Now I have a question of how to create a map (occupancygrid map) and fuse the position to robot_localization package.

If you look inside node.cc here LINK you can see how the position is being extracted from frames (correct me here if I am wrong) and TF between MAP and Camera is being created.

Will I be wrong here if I:

1. Remove the map frame
2. Get that TF above and fuse as odometry message to robot_localization package as a global sensor
3. Use point cloud to laser converter
4. Use converted point clouds as the input to gmapping and perform SLAM.
5. orb slam has a localizer but I guess it works with point clouds so I can use that on the future instead of AMCL.