I recently extended <https://www.openstreetmap.org/way/4072429#map=18/52.22564/0.11783&layers=N> some already mapped pavements in N. Cambridge. I'm not really a fan of the current approach because I don't think it works particularly well, and I'm not aware of any good routers using this type of data for wheelchairs.
The problems I see (and I've said this before): - The scope for missing interconnections is trebled. - It's more or less worthless unless done systematically (places like university & hospital campuses are viable from this viewpoint. - In Britain, at least, it requires introduction of many arbitrary crossing points to allow any kind of sensible pedestrian routing (i.e., not well-supported by on-the-ground features such as dropped kerbs & tactile paving). You can see the ones I felt it necessary to introduce around Roseford Road & Perse Way. Note that many crossings, e.g., at the Harris Way/Perse Way intersection are not complete. - It breaks existing applications. The reason why I noticed the issue in North Cambridge is that the Traveline South East app started giving me unfeasibly long times to walk to a bus stop. It turned out that it routed me all the way along a pavement to Histon Road & then back along Histon Road adding a good 500 m to the journey. This was because the original mapping just stopped without connecting the end of the pavement to anything. - I'm not completely convinced that wheelchair users, blind people etc can put the same degree of trust in this type of data as the ordinary pedestrian can for current pedestrian routing. My feeling is that the information really needs to be tailored to the user: there's a massive difference between how a powered wheelchair or mobility scooter and a manual/pushed wheelchair can cope with non-flush kerbs for instance. - I'm not sure if anyone has done any work to show how separately mapped sidewalks can be merged with the main highway to provide generalised pedestrian routing such as we have now. - Probably to be useful in the UK, all driveways should be mapped too (as in Andy's dev server example): in my experience of pushing my late mother around in a wheelchair driveways are often much better than many shoddy dropped kerb installations. - Naming of sidewalks can create problems (although it can also resolve them in cases where the two sides of a street have different names). - It's a pig to survey well in places where dropped kerbs have not been installed systematically (as in my Cambridge example). On the plus side: - It allows more relevant details of pavements to be tagged (width, surface etc). - The current sidewalk model is probably much more appropriate in countries with specific legislation preventing pedestrians crossing roads at any other than designated crossing points (jay walking). - It's always been good publicity for OSM: even if actual real usage is limited. - Inevitably OSM will move in the direction of capturing more information & this is just one example. I guess I would have preferred : sidewalks to be mapped with a key other than highway (something analogous to area:highway); more research to be done on ways to post-process the data (in both directions from highway=footway,footway=sidewalk and from sidewalk=*); and good references for actual user experience of wheelchair routing using separately mapped sidewalks. One way to have our cake & eat it would be to use both sidewalk= and have separately mapped sidewalks & allow the consumer to choose which to use, although the current sidewalk=separate does not say if its both, left or right. Personally I think this is still reasonable in the context of one feature one element; sidewalk is an attributive property of the street and potentially difficult to derive without resorting to convoluted approaches (such as relations). In summary the problem from my perspective is that mapping them separately can often make OSM less useful, whereas most other mapping of additional features enhances OSM incrementally. Jerry On Sat, 1 Jun 2019 at 15:08, Andy Townsend <ajt1...@gmail.com> wrote: > On 01/06/2019 13:55, Michael Collinson wrote: > > > > ... I tried, then going out to "just verify" and found that I was > > hopelessly inaccurate. It defeats the point, to get a highly accurate > > localised network for folks who might depend on it. > > > > > I did something similar on the dev server a while back here: > > https://master.apis.dev.openstreetmap.org/#map=16/54.0167/-1.0486 > > (turn the data layer on to see it). What surprised me was the things > that I hadn't expected beforehand to be important (angles through gates > being an obvious one) that actually were. > > Best Regards, > > Andy > > > > > _______________________________________________ > Talk-GB mailing list > Talk-GB@openstreetmap.org > https://lists.openstreetmap.org/listinfo/talk-gb >
_______________________________________________ Talk-GB mailing list Talk-GB@openstreetmap.org https://lists.openstreetmap.org/listinfo/talk-gb
