1
$\begingroup$

I have two bag files with significant overlap and multiple loop closure opportunities. However, these bag files were recorded at different times. I want to use these bag files to reconstruct a scene and chain the reconstructions together. Currently, I am using fastlio with scancontext.

However, when I run the second bag file in the same fastlio session, the reconstruction breaks, possibly due to imu initialization issues. If I keep the scancontext session and restart the fastlio session, the reconstruction gets placed at 0,0,0, and scancontext cannot find a proper match.

Can anyone suggest a workflow for chaining the reconstructions together using multi-session data and bag files with significant overlap? How can I avoid these initialization issues in fastlio and maintain the proper placement of the reconstruction using scancontext?

I would also like to ask if there are alternative solutions for using Livox LiDARs (Mid-360) and offline SLAM to chain reconstructions together with multi-session bag file data.

$\endgroup$

2 Answers 2

1
$\begingroup$

After struggling with this for a while, the best repo I have found for offline reconstruction/pose graph optimization and merging multisession data is interactive slam. https://github.com/SMRT-AIST/interactive_slam

$\endgroup$
0
$\begingroup$

Have you tried the graph optimization approach?

  1. Cut the fastlio local map into segments
  2. Generate a pose graph from the segments
  3. Repeat 1,2 for the other dataset
  4. Run a place recognition algorithm like BoW3D or scancontext to find the correspondence between the segments
  5. Find the 6 dof transformations between the matchings
  6. Add them to the graph
  7. Run the optimization
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.