Loop closing

[MaEppl]

Hey,
I'm trying to use ethz-asl/orb_slam_2_ros + voxblox. I think the ros version of orbslam only publishes the current camera position. How about loop closing, how can I get the final trajectory? On my quadrocopter I would like to see the actual position. But afterwards the reconstruction should be based on the optimized trajectory.
Thanks!

[MaEppl]

@alexmillane I saw that you've already worked on "loop closure trajectory publishing". Why have you reverted your commits on the master branch, in your personal repository it seems like you are using those functionalities?

[ZacharyTaylor]

From memory the work on exposing the full trajectory and other info was done for the sub-mapping approach. However the sub-mapper needed a large amount of two way communication and was causing too much coupling between the nodes. Because, of this the sub-mapper was refactored to include a modified version of orb-slam inside it, which is why the exposing of the orb-slam internals to ros wasn't completed / was reverted.

[alexmillane]

Correct Zach. For a while I worked on publishing a lot of information from orb_slam_2 over ROS but in the end it got too much and I decided just to use it as a library for what I was using and access the internals directly.

The problem is that orb_slam_2, by default, does not expose an interface for getting an optimized trajectory back (or anything other information). The only information it provides is the current tracking pose. Because I wanted this interface to rely on orb_slam_2 master, that's what we're stuck with.

If you want to get the trajectory, have a look through what I did on my personal forks, as it appears you already have.