Structureless VIO

[Martin Ennemoser]

Hello.
I want to build a VIO pipeline whereas the landmark positions are known and dont need to be optimized.  I want to use iSAM2 for that. I read in this paper https://arxiv.org/pdf/1512.02363 about structureless model, where landmark positions are marginalized out of the state vector. How can I do that in GTSAM? 

[Dellaert, Frank]

Hi, Martin,

Your scenario is localization, and it is great use for ISAM2, but structureless factors are actually a concept in SLAM. You're still estimating the landmarks there. 

My answer on how to do this would depend on what language you're trying to do this in. It's in Python or MATLAB, probably creating a Pose3 (or NavState) custom factor for each timestamp that evaluates the re-projection error for all known landmarks seen from that timestamp in parallel is going to be the fastest and most convenient.

Frank

--
You received this message because you are subscribed to the Google Groups "gtsam users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to gtsam-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/gtsam-users/93cf196d-ac11-45dc-a076-b344ebc02318n%40googlegroups.com.

[Martin Ennemoser]

Hi Frank

thank you for the answer. Maybe I mistake some concepts. So I want to do localization in Python using a camera and an IMU. Can you give some guidance or some Python example code how to do it? That would be very helpful.

Bests, Martin

[Dellaert, Frank]

Not exactly, but if tell Codex what you want and also give it the guidance I provided below , it will probably do a pretty good job. 

To view this discussion visit https://groups.google.com/d/msgid/gtsam-users/fb7afc54-7ccb-442e-a6ae-aba501399ea1n%40googlegroups.com.

[Martin Ennemoser]

Thank you. I give it a shot.

[Dellaert, Frank]

All

I just posted a new blog post on gtsam.org about the CustomFactor. I also added two examples, one of which was inspired by Martin’s question: https://borglab.github.io/gtsam/customfactorlocalizationexample/ 

Martin, this is is simpler than a projection-based localization, but should give you a good starting point 😊

Best

Frank

To view this discussion visit https://groups.google.com/d/msgid/gtsam-users/31179f16-2291-4ed2-b3bf-dcd9297d0997n%40googlegroups.com.

[Martin Ennemoser]

Thank you very much for your effort.

One dumb question because I am still not an expert in GTSAM: The landmarks are not optimized right?