Best way to contribute inferred roads from delivery traces?

Help and support
, ,
1

Hi everyone,

I’m reaching out from Delhivery pvt ltd. Our delivery fleet travels across a massive area of India every day, and we’ve noticed that many of the local roads, residential streets, and paths they use aren’t currently mapped on OpenStreetMap. We want to give back to the OSM community by making this data available to help fill these gaps!
Here is where we are currently at: We have taken our raw, anonymized GPS delivery traces and used map inference algorithms to denoise the data, successfully generating clean, suggested road centerlines.
However, we want to make sure we contribute this properly. We know that automated, direct imports are strictly against OSM guidelines, and we are absolutely committed to a “human-in-the-loop” approach to ensure data quality and correct topology.
Before we finalize a workflow for our internal mapping team (and hopefully community volunteers!), we want to understand your POV. We would love your guidance on the following:

  1. What tools or platforms do you recommend? How does the community prefer to review and validate this kind of generated geometry against satellite imagery?
  2. Project Structure: How can we present this data so it is genuinely useful and easy for local mappers to work with, rather than being an overwhelming data dump?
    Tagging:
  3. Are there specific tagging conventions you’d recommend for roads inferred from delivery routes (e.g., primarily two-wheelers)?

We want to align our process with how OSM community actually likes to work. Any advice, tool recommendations, or feedback would be immensely appreciated!
Looking forward to hearing your thoughts.

4 Likes
2

Hello,

thanks for wanting to give back to OSM. That seems like a really good way to support and contribute to OSM in general. I will try to answer to the best of my knowledge, but i will summon @contrapunctus here as well, as they are from india if i recall correctly :)

We know that automated, direct imports are strictly against OSM guidelines, and we are absolutely committed to a “human-in-the-loop” approach to ensure data quality and correct topology.

Thats actually not completely correct, but some guidelines needs get followed. They can be found in the wiki Automated Edits code of conduct - OpenStreetMap Wiki
For organised editing (also, company editing) there are the Organised Editing Guidelines - OpenStreetMap Foundation as well.

But, with opening a thread in the forum you made a good first step for both cases - opening a discussion.

So, uploading the GPX-Traces on osm.org would be a good first step, there are apps for smartphone that are doing it automatically, there are also projects like OpenPilot or SunnyPilot that automatically upload all their driven ways to osm.org as far as i know :)
The more GPX routes there are, the better a street can be drawn or the aerial imagery can get corrected. I don’t say you should use this hardware, but uploading the GPS-Traces are already a big help and a good start.

For you point 3 - how to tag the roads/highways/… i would suggest reading through the wikipage about highway Key:highway - OpenStreetMap Wiki
Bascially every road can additionally tagged with Key:surface - OpenStreetMap Wiki and or Key:smoothness - OpenStreetMap Wiki to indicate the quality of the road.

But, i think contrapunktus can help as well :)

3

presumably you want to avoid sharing GPS tracks themselves, right?

2 Likes
4

Yes, exactly — we are cautious about sharing raw GPS traces directly because of both privacy and operational sensitivity concerns.

5

The data overlays in RapID / JOSM MapWithAI seem to be effective ways of presenting “missing road line” data from what I’ve heard. Don’t know what’s involved to do data conversions there.

6

From what I currently understand, RapiD does not directly support plugging in arbitrary custom datasets into the MapWithAI suggestion workflow.

Custom datasets can be loaded as overlays/reference layers, but they are not treated as editable AI suggestions like the built-in Meta/TomTom layers.

I may still be missing some integration mechanism internally, but so far it seems that deeper MapWithAI integration is required rather than simply exposing a GeoJSON/vector overlay.

1 Like
7

Hi @Harshit1304 ! That’s a good news, thanks a lot for contributing back to OSM. Some comments:

  • have you consider contributing to Panoramax ?
  • how do you manage to get an accurate precision in some very dense area ? In Delhi, in very dense area, you can’t distinguish road from aerial imagery, and the gps is extremely not precise
  • not sure if it’s the best practice, but this is what I’m using to map very narrow road Key:maxwidth:physical - OpenStreetMap Wiki . Lots of very narrow road are tagged with highway=service and when I’m driving with my car, OsmAnd wants me to take a shortcut which is suitable only for tuktuk / bike. How do you plan to map those roads ?
8

You could remove the first and last couple of 100 metres (or other suitable distance) from the tracks prior to upload. If this makes sense will depend on many factors, for example trip length etc.

More generally, India is a large country and the OSM community is still quite small which limits what can be done with your data (you will still need road classification etc. besides the raw geometry), maybe the better approach would be to support the community in growing further and more rapidly.

4 Likes