You might find @Dustin_C‘s How did the mapper cross the road? presentation at SoTM interesting on permeability/severance
2 Likes
An example of routing before adding “separate” mapped pavements, even when they are attached to the road, and after when a separate one has been drawn.
OSM Routers would prefer you go down Jack the Rippers alleyway vs the more suitable route.
Route Before:
Route After:
This is what it is like everywhere.
The very idea that pedestrian routing “already worked fine before” is 100% uninformed and false.
That’s not actually what I’m seeing in the OSM data. The history of the middle bit outside the cricket ground suggests that it didn’t have any sidewalk tags at all until a couple of days ago. The before/after comparison is therefore between “no sidewalk at all” on the A5095 and “a separately mapped one”. There may well be routers that show results like yours, but this isn’t an example of one. If there are, it’d be a bit niche preferring e.g. highway=footway; footway=sidewalk over highway=<some sort of road>; sidewalk=both, given that this two situations represent exactly the same on-the-ground situation.
Quite the reverse, actually - there are a vast number of different variables that could potentially affect routing, so everywhere is different. To compare before and after for rendering I tend to use timestamped copies of data from Geofabrik (see the “raw directory index” on this page), usually on two servers, but given the processing involved for routing that’d actually be a bit of a challenge. It’d be interesting to hear what other people do.
3 Likes
This might not be the case for GraphHopper and the other demo routers on osm.org, but some others might add a higher cost to the alley if it was tagged with lit=no + surface=sett + smoothness=intermediate|bad. I believe TrailRouter (for running routes) has an option to avoid unlit roads and paths.
That’s not an argument against mapping separate sidewalks on Abington Road - I would probably add them if it was near me.
2 Likes
Not to pile on but yeah, what you’re seeing here isn’t an advantage of adding a separated pavement, it’s just an unintended consequence of adding a pavement to the road at all.
Abington Avenue is on the edge of what I would map separately, and would depend both on how busy the road was and how often there were low/flush kerbs. The presence of a pelican crossing on the road would imply that it is “unevenly permeable” (the pelican crossing being more permeable than the rest of the road), which in turn would support a separated pavement.
A quick wander identifies Roe Road as somewhere I would say looks “evenly permeable” (you can cross pretty much anywhere, with many dropped kerbs and a generally low kerb-line) and as such I wouldn’t put a separated pavement there (because it doesn’t add any new information).
(Facebook would probably map sidewalks on the alleyway..)
1 Like
This was the point of the experiment. to take a normally mapped road, and show the failures of relying on roads for pedestrian mapping. If pedestrians are going to go into that much effort having to re-work the road network to cater to pedestrian infrastructure, we may as well add separate sidewalks when the data is appropriate.
Because if we are going to start suggesting not mapping separate pedestrian infrastructure because “sometimes it is left unfinished” we need to look at the black hole sized gaming hole in the dataset of how incompletely mapped roads are.
Northampton has just 490 ways tagged with sidewalk=yes, sidewalk=both, sidewalk:both=yes, sidewalk:left=yes, or sidewalk:right=yes
That is a town with nearly a quarter of a million people.
Kettering has 36,000 people, but 1,300 ways tagged the same way. Per capita that is 18 times more sidewalk tags in a place that has had effort to map the sidewalks, vs a regular mapped place.
Should we be putting all road mapping on hold if the mappers can’t accurately tag the state of the sidewalks when they map the roads?
The numbers go up to 2,145 ways with sidewalk tags in Kettering if you include separate tags, and 1,435 in Northampton (the vast majority being along a primary corridor into the town).
It is also in response to examples where the pavements have had correct tags on the road, for years, and no router ever picked up the correct side of the road you should be on, instead having you cross at a cycle-way exit from a 3 lane wide road, to then cross back to the side you were already on at a pedestrian crossing
Right, so if the town hasn’t even been mapped properly with the “simple” approach of using a single way for a road (and then tagging pavements, carriageways, etc on that way) then how are separated pavements helping with this?
Do whatever makes you happy, but if you’re trying to improve the experience for pedestrians in your area then walking around with StreetComplete filling in all the pavements (accurately) would have a hell of a lot more impact than all the effort it would take to properly survey and then draw out all the pavements separately to the roads.
This was really interesting, his idea of “severance” (do you need to think about crossing the road) is similar to what I’m saying with “permeability” (is the road mostly crossable at any point). It’s surprising to me, with that in mind, that he proposes “sticking to one or the other” in terms of pavement mapping techniques instead of representing severance with separated pavements even where they’re not otherwise used.
I also can’t help but feel that something like speedwalk makes separated pavements redundant in areas of simple geometry - if we had a severance=yes|no tag for roads then that would represent severance such that a router could automatically generate the pavements and then route across crossing nodes.
Good talk, anyway.
This table never gets old. In the line “Detailed accessibility tagging”, we have that “tag on highway” is “usually sufficient, but not when most needed in complex spaces”. So, for streets and pedestrian crossings at intersections that follow the most common pattern (continuous sidewalks on both sides, crosswalks or implicit allowed pedestrian crossings on all four corners of intersections with kerbs and no ramps), the simplest possible mapping (using only tags) can represent reality sufficiently. If you need to represent something different from that (for example, the presence of ramps, or the absence of a crosswalk on one side side), then yes, the “separate footway” method is the only way to accurately represent it.
My area is quite heterogenous, but it’s fair to say that the vast majority of local (residential) ways follow the most common pattern, as do many main ways (especially tertiary ways and many secondary ways). Adding separate sidewalks for all of them would be overkill and maintenance hell. I personally adopt both mapping styles, using tags for simplicity where possible, and separate footways for detail where necessary.
3 Likes
@LordGarySugar I can’t help but feel like this changeset, which adds separated pavements specifically around the area I’ve been mapping in detail, is specifically targeted to wind me up - it does unfortunately also have some errors in it (the aerial photography is old) and it does have some other opinion-based edits in it (such as moving land-use back from the road way to the cadastral boundary, which I know we disagree on from another thread).
Where do we go from here? I’m not going to get in an edit war with you. I have been very careful to raise potentially controversial changes (such as the above in Manchester city centre) on the forums first, I’ve put comments on my changesets to link back to the discussion, and once it became clear that there was resistance to my proposals I backed off making any further changes.
I understand you’re a prolific mapper but this feels wrong.
Isn’t it, though? The push now is for street:name=* not name=* on footways, and “Consideration: Routing is continuous at intersections with unmarked pedestrian crossings” + “Separate footway: Requires mapping informal crossings or footway links to prevent routing islands/detours” makes no sense as a combination, because (as the “Consideration” part literally says) those are unmarked crossings (crossing:markings=no).
They’re not unmarked crossings. You could possibly argue that there is a de facto unmarked crossing at the end of any given road, but not e.g. across the continuing road, or where a road meets a footpath, a shopping mall entrance, a train station entrance, across alleyways that bisect neighborhoods, etc etc.
The point the table seems to be making is that you need to draw in “phantom" unmarked crossings in all these places to ensure routing doesn’t take you round the houses.
1 Like
That doesn’t make sense, unless you believe that sidewalks are created only for the purpose of routing around a single block. That’s of course not true - it’s intended that pedestrians are able to cross.
1 Like
Even where separate sidewalks aren’t mapped, I’d still map an unmarked crossing at e.g. a shopping centre entrance if there is a lowered kerb, or tactile paving, or similar.
1 Like
Who is pushing this? Is this organized editing? (TagInfo does tell me that the number of occurrences of street:name exploded since 13 January 2026. Apparently, it was around this time that OSRM began to support this tag.)
It still makes sense, undoubtedly. In many jurisdictions, crossing at an intersection is permitted even if there are no marked pedestrian crossings. If the links mentioned in the wiki are missing, the routing system will fail to consider perfectly valid pedestrian movements. The absence of these links often causes nonsensical detours where sidewalks are mapped separately from the main way.
However, this situation likely requires further discussion, since if there is a marked pedestrian crossing slightly further away from the intersection, the unmarked crossing at the intersection should be heavily penalized by the routing system to prioritize the marked one. There is probably a limit to the expected extra walking in this situation.
Is it still early enough to correct course on a new tag’s name? The colon ( : ) is usually intended for subtags, but street is not a tag, so this tag should probably be called street_name, not street:name. street could be a “tag aggregator” like addr, but there doesn’t seem to be any other tags prefixed by street right now.
Yes, what I mean are crossings placed to emulate the fact that you can cross anywhere on the road, which you need to include to avoid the expedition half way up the street, and are generally not crossings by any meaning of the word - that is, nothing is lowered or tactile, and people are just as likely to cross there as any other point on the street.
The table could be clearer but my reading of it is a comparison with the “natural" continuity an integrated pavement gives you.
I think the US mappers are going with this but it’s certainly not universal - in Canada at least they’re using is_sidepath:of:name and I believe London is doing the same thing.
Of course, if the US had their way..
1 Like
In London, it’s probably down to me. I’d already been adding name for quite some time. When the is_sidepath:of:name proposal came along, I started using that in parallel. After street:name appeared (or I was made aware of it), I’ve used that instead, alongside name. I don’t really have a strong opinion about which of the three ways of recording street names on separate sidewalks is best and will happily adopt whichever eventually becomes the de facto standard.
1 Like
I was editing in the Manchester area and noticed you had been editing nearby so I had a nose around at what had been mapped. I’d like to make clear that I was not only adding pavements, but re-adding pavements that you yourself deleted, without justification. (See Changeset: 173575115 | OpenStreetMap ). It’s all well and good to make the argument for deleting incorrect or nonexistent pavements, but this doesn’t track with your edits where you have removed correctly mapped sidewalks. Plus, I corrected a lot of the existing sidewalks and crossings so hopefully there is more useful pedestrian detail now, although there could probably be more informal crossings added on residential roads.
If you can point me to where the mapping is incorrect for this reason I’d be happy to correct what I can, but without any information about what those errors might be I can’t respond to that.
This is not an opinion-based edit, it is not correct to map landuse extending to the centre line of roads, in the same way that gluing landuse to roads is not correct.
If you point to some specific areas you think need improvement, I will gladly add some more details or informal crossings, but I am not going to engage with any more ‘separated sidewalks bad’ discourse. I did make an effort to make sure all the pavements were as detailed as possible with kerbs in the right place and everything. As far as I’m concerned, everything is mapped as best as I could and is in no way controversial as you seem to imply it is.
2 Likes