Last weekend I joined an in-person workshop at HTW Berlin for discussing topics around mapping indoor spaces in OpenStreetMap.

Indoor mapping

Mapping indoor spaces is a somewhat niche topic in the OSM community still, but something that is quite relevant for projects I’m involved in:

The term “indoor” isn’t strictly referring to “in a building” here, there’s many gray areas e.g. at train stations. What’s usually more important is that this introduces a third dimension into the originally mostly two-dimensional OSM data. Another challenge is the need for a very high level of detail, for wheelchair routing every single step matters for example.

Since the last workshop four years ago we had quarterly online meetups to discuss modeling and tagging questions, but for some topics an hour or two in an online meeting is just not enough to properly cover this, it needs locking people in a room for a day or two instead.

Balancing requirements

While it’s often not hard to find a solution for a modeling problem at hand, finding one that works well for all use-cases is hard:

  • Easy to explain, use and maintain. Don’t require deep domain knowledge of railway operations or civil engineering for mapping.
  • (Backward) compatible with existing data and existing semantics. OSM is a database with billions of objects built up over more than two decades, with countless users.
  • Usable for 2D rendering.
  • Usable for 2.5D or 3D rendering. This specifically benefits from a higher spatial resolution in the third dimension (which currently is essentially floor levels, compared to the centimeter resolution in the other dimensions), but also from the ability to map visuals of vertical features.
  • Usable for tactile rendering. This one is challenging as it’s actively harmed by a higher level of detail. On dynamic tactile displays you have to work with as little as 100 “pixels” in each direction, and possibly just one 1 bit of “pixel depth”, which requires very aggressive abstraction and simplification. 3D printed tactile maps have a higher spatial resolution and allow for some basic textures and symbols, but still way below the options you have with a visual display.
  • Usable for routing, both with graph-based and area-based algorithms. The main challenge here is completeness of the data, every barrier has to be there to avoid the router taking clever shortcuts.
  • Being able to model all accessibility-relevant properties. See the discussion on directional door attributes below, for example.
  • Compatibility for importing BIM data, ie. digital engineering models of buildings. That’s on the extreme end of the level of detail usually, and needs to be significantly reduced/simplified. But it’s an attractive source of high quality building geometries for say an entire university campus.

Fortunately we had people familiar with all those aspects present at the workshop, which helps to avoid easy one-sided compromises.

What even is a door?

There’s detailed notes on the discussion in the wiki, I’ll just pick one topic here to show how even seemingly easy and obvious things are surprisingly complicated when digging into the details, doors.

  • Is a door frame with the actual door removed still a door?
  • Is a multi-segment foldable door as e.g. found in shop fronts or between conference rooms a door? If so, does that mean doors can contain doors?
  • Doors are 0-dimensional features (ie. points) in OSM. That’s a useful simplification, but how do we define directional attributes in that case, e.g. in which direction does the door open?
  • Inside/outside is an intuitive way to do that, but that fails in more complex buildings. It would also mean evaluating door attributes requires determining an “inner-ness” hierarchy of all areas in a building, extremely unwieldy, if that would even be well-defined.
  • Modeling doors as 1-dimensional features (ie. lines) brings in an inherent direction, but that is perpendicular to the intuitive direction of standing in front of a door, requiring some mental gymnastics to make this work (“I am the door”).

This might seem pointlessly abstract and theoretical, but for e.g. wheelchair routing this is quite relevant. The opening mechanism and opening direction of a door have quite some impact on how easily you can get through.

Outlook

State of the Map 2026 logo

Indoor mapping will probably also be a topic at State of the Map 2026 in a few weeks in Paris. I’ll be speaking about Transitous there, one of the consumers of this data.