Introducing DifferentialPseudorangeFactor

[masoug]

Hello again! I've experimented with applying differential corrections to the PseudorangeFactor and found ~98% accuracy improvement. I've attached a copy of my experiments here: DifferentialPseudorangeExample.html

Similar to before, I'm seeking feedback on:

• How to structure the PseudorangeFactors to handle commonality b/w PseudorangeFactor and DifferentialPseudorangeFactor? So far I went with simple inheritance, but if there's a more elegant way to arrange the classes, let me know!
• What are preferred conventions for notebook library modules? I placed a gnss_utils.py to de-duplicate a lot of repetition, so I'm wondering if that's okay for the two pseudorange examples.
  • Would gnss_utils.py cause issues with google colab?
• DifferentialPseudorangeExample.ipynb; what could be clarified, and what are some suggestions to communicate the significance of going from 30->0.4 meter accuracy? Does anyone know of good geospatial map plotting libraries that work well with Jupyter? ChatGPT suggested folium...

Anyway, I'm interested to hear what y'all think 🧠 I think I might look at carrier-phase factors next, so @varunagrawal I'd be interested in learning more about your experiences on that, if you're willing to share 🙏

Best,
Sammy

[ProfFan]

Thank you for the contributions!

How to structure the PseudorangeFactors to handle commonality b/w PseudorangeFactor and DifferentialPseudorangeFactor?

I would recommend splitting out the common stuff as standalone utility functions instead of inheritance

What are preferred conventions for notebook library modules?

I think you can put things in gtsam_unstable but then it won't work until the next nightly build. For now a simple folded cell probably works better?

good geospatial map plotting libraries

https://docs.kepler.gl/docs/keplergl-jupyter

[dellaert]

Very nice. Your factor graph illustration is also great, but maybe it would be more clear if you showed the correction and the pseudo factors for two satellites. You really have to read the text to understand that there one correction variable per satellite.

Also wondering what happens over time and as the receiver moves.

[dellaert]

Yes, I walked in them (Coyote) hills. There used to be nuclear missiles stationed there, too 🚀

[dellaert]

https://en.wikipedia.org/wiki/Project_Nike

[varunagrawal]

I don't think putting code in gtsam and the notebooks in gtsam_unstable is a good idea. It will make finding each part tricky.

I have used folium in the past and it is great! It makes really nice interactive widgets that one can play around with, and the API is really simple to get basic map and path rendering.

[dellaert]

Agreed. @masoug, please put that module in the examples folder, next to the notebooks, and do the wget on colab.

[masoug]

Thanks for the feedback! I'll get these comments addressed in the next couple days 👍

[masoug]

Very nice. Your factor graph illustration is also great, but maybe it would be more clear if you showed the correction and the pseudo factors for two satellites. You really have to read the text to understand that there one correction variable per satellite.

I've updated the illustration by adding an image (I couldn't get mermaid render a decent graph) showing two satellites and their pseudorange factors. Attached is an updated notebook output: DifferentialPseudorangeExample.html

Also wondering what happens over time and as the receiver moves.

This is a great question! Spoiler alert: Position quality degrades over long times and great distances between the user and reference stations. I wanted to run an experiment where I plot accuracy as a function of distance-from-reference-station for a variety of test station across CORS, but I'd put that in a separate notebook not designed for documentation/tutorial.

[dellaert]

Awesome ! cool new graphic :-)