I investigated on this idea a bit further. I tried to arrive at numbers that when multiplied would make sense to me, but in the end your suggestion of speed limits for each tag made more sense. So, instead I would now agree with taking the minimum of the following speeds: default, max, tracktype limit, smoothness limit, and other things. And I have a new suggestion: we can try to guess tracktype and smoothness from surface, but only do so when these tags are missing.
I've come to this conclusion after learning what tracktype means exactly: https://lists.openstreetmap.org/pipermail/tagging/2014-March/016904.html I've also asked the opinion of the community on this idea: https://lists.openstreetmap.org/pipermail/tagging/2014-March/016935.html I'd like to hear your opinion about adding the following code to https://github.com/DennisOSRM/Project-OSRM/blob/master/profiles/car.lua: 1. Just after the definition of speed_profile (line 34): ----- tracktype_profile = { ["grade1"] = math.huge, ["grade2"] = 45, ["grade3"] = 30, ["grade4"] = 20, ["grade5"] = 15, ["grade6"] = 9, ["grade7"] = 6, ["grade8"] = 3 } surface_tracktype_profile = { ["asphalt"] = tracktype_profile["grade1"], ["concrete"] = tracktype_profile["grade1"], ["tartan"] = tracktype_profile["grade1"], ["paved"] = tracktype_profile["grade1"], ["paving_stones"] = tracktype_profile["grade1"], ["concrete:plates"] = tracktype_profile["grade1"], ["metal"] = tracktype_profile["grade1"], ["compacted"] = tracktype_profile["grade1"], ["sett"] = tracktype_profile["grade1"], ["concrete:lanes"] = tracktype_profile["grade1"], ["bricks"] = tracktype_profile["grade1"], ["cement"] = tracktype_profile["grade1"], ["cobblestone"] = tracktype_profile["grade1"], ["wood"] = tracktype_profile["grade1"], ["stone"] = tracktype_profile["grade1"], ["rocky"] = tracktype_profile["grade1"], ["gravel"] = tracktype_profile["grade2"], ["fine_gravel"] = tracktype_profile["grade2"], ["grass_paver"] = tracktype_profile["grade2"], ["unpaved"] = tracktype_profile["grade3"], ["ground"] = tracktype_profile["grade3"], ["dirt"] = tracktype_profile["grade3"], ["grass"] = tracktype_profile["grade3"], ["pebblestone"] = tracktype_profile["grade3"], ["clay"] = tracktype_profile["grade4"], ["sand"] = tracktype_profile["grade5"], ["earth"] = tracktype_profile["grade5"], ["mud"] = tracktype_profile["grade5"] } smoothness_profile = { ["excellent"] = math.huge, ["thin_rollers"] = math.huge, ["good"] = 60, ["thin_wheels"] = 60, ["intermediate"] = 45, ["wheels"] = 45, ["bad"] = 30, ["robust_wheels"] = 30, ["very_bad"] = 15, ["high_clearance"] = 15, ["horrible"] = 3, ["off_road_wheels"] = 3 } surface_smoothness_profile = { ["asphalt"] = smoothness_profile["excellent"], ["concrete"] = smoothness_profile["excellent"], ["tartan"] = smoothness_profile["excellent"], ["paved"] = smoothness_profile["good"], ["paving_stones"] = smoothness_profile["good"], ["concrete:plates"] = smoothness_profile["good"], ["metal"] = smoothness_profile["good"], ["compacted"] = smoothness_profile["intermediate"], ["sett"] = smoothness_profile["intermediate"], ["concrete:lanes"] = smoothness_profile["intermediate"], ["bricks"] = smoothness_profile["intermediate"], ["cement"] = smoothness_profile["intermediate"], ["grass_paver"] = smoothness_profile["intermediate"], ["cobblestone"] = smoothness_profile["bad"], ["wood"] = smoothness_profile["bad"], ["stone"] = smoothness_profile["bad"], ["rocky"] = smoothness_profile["bad"], ["gravel"] = smoothness_profile["bad"], ["fine_gravel"] = smoothness_profile["bad"], ["unpaved"] = smoothness_profile["bad"], ["ground"] = smoothness_profile["bad"], ["dirt"] = smoothness_profile["bad"], ["grass"] = smoothness_profile["bad"], ["pebblestone"] = smoothness_profile["bad"], ["clay"] = smoothness_profile["bad"], ["sand"] = smoothness_profile["bad"], ["earth"] = smoothness_profile["bad"], ["mud"] = smoothness_profile["very_bad"] } ----- 2. At the beginning of way_function(), just after checking for access (line 119): ----- -- we dont route over extremely difficult surfaces local surface = way.tags:Find("surface") local tracktype = way.tags:Find("tracktype") local smoothness = way.tags:Find("smoothness") -- accept only widely used tracktype values (typos result in inaccessible ways) if tracktype ~= "" then if tracktype_profile[tracktype] == nil then return end end -- accept only widely used smoothness values (typos result in inaccessible ways) if smoothness ~= "" then if smoothness_profile[tracktype] == nil then return end end ----- 3. Just before handling forward/backward maxpeeds (line 206): ----- -- Set the avg speed on ways with difficult surfaces if tracktype ~= "" then way.speed = math.min(way.speed, tracktype_profile[tracktype]) else if surface ~= "" then if surface_tracktype_profile[surface] ~= nil then way.speed = math.min(way.speed, surface_tracktype_profile[surface]) end end end if smoothness ~= "" then way.speed = math.min(way.speed, smoothness_profile[smoothness]) else if surface ~= "" then if surface_smoothness_profile[surface] ~= nil then way.speed = math.min(way.speed, surface_smoothness_profile[surface]) end end end ----- On Mon, Mar 10, 2014 at 11:11 PM, Fernando Trebien <fernando.treb...@gmail.com> wrote: > "But if we do want to handle combinations" > > Shouldn't we handle them if the community thinks that surface is only > fully described after adding the 3 tags? Nowhere it is said "use one > or the other but not both". > > The very fact that people don't often combine them means to me that > one is simply a rough copy of the other and each is adopted by a > different community. > > "maybe use surface (the most used tag), multiplied by a factor if > tracktype or smoothness is set as well?" > > What if both are present? If we multiply by both factors, the final > result would be lower than expected. In that case, at least averaging > them would be better (if you're looking for a very simple method). > > "In essense, take minimium of: > - default speed > - max speed > - surface speed * tracktype or smoothness factor > - other things that lower the speed" > > Makes sense to me. > > On Mon, Mar 10, 2014 at 3:10 PM, Emil Tin <e...@tin.dk> wrote: >> The tags are not used so often together (from >> http://taginfo.openstreetmap.org/keys/surface#combinations): >> >> Surface (8 077 811 occurences) combinations: >> 1 087 838 13.47% tracktype >> 175 011 2.17% smoothness >> >> Tracktype combinations: >> 46 144 1.44% smoothness >> >> But if we do want to handle combinations, maybe use surface (the most used >> tag), multiplied by a factor if tracktype or smoothness is set as well? >> >> In essense, take minimium of: >> - default speed >> - max speed >> - surface speed * tracktype or smoothness factor >> - other things that lower the speed >> >> >> >> On 10 Mar 2014, at 17:30 , Fernando Trebien <fernando.treb...@gmail.com> >> wrote: >> >> For each tag+value combination, we could assign: >> - a preference value >> - a weight value or a priority value >> >> When using weights, the final preference value would be a weighed >> average of the preference values corresponding to each tag. >> >> When using priorities, the final preference would be that of the >> highest priority tag+value combination. >> >> Contradictions are an issue, for example: >> - tracktype=grade1 + smoothness=very_horrible: is it good or is it bad? >> - tracktype=grade5 + surface=asphalt: is it paved or not? >> - smoothness=impassable + surface=concrete: maybe the concrete path >> was very badly built, or maybe we just had an earthquake >> >> Weighting would "blur" the contradiction, opting for an average >> preference of the conflicting values. The greater the contradiction, >> the greater the risk of poor routing decisions. >> >> A pessimist approach (probably "safer" for the user) is to select the >> lowest preference value assigned for tag+value combinations present in >> a way. But then, we lose the ability to use one tag as a refinement >> for the other (for example: tracktype=grade2/3/4/5 when >> smoothness=bad). >> >> A remedy would be a more complex approach which practically encodes >> the class system that I proposed: >> - given 3 tag+value combinations, pick the combination with lowest >> preference value (let's call this tag A) >> - for the remaining 2 combinations, select those that are considered >> similar to A (according to some equivalence table) and discard the >> others >> - if there are 2 left, also pick the one with lowest preference value (tag >> B) >> - possibly pick tag C if it's similar to both A and B >> >> The result would be from 1 to 3 tags (A,B,C) from which you'd choose >> the one with highest priority. That's the most accurate preference >> value within pessimist choices. >> >> A single classification system would eliminate these problems, and it >> can be introduced in the community (not necessarily in OSRM) >> simultaneously with a more temporary solution in OSRM using multiple >> tags and some sort of contradiction handling. I just wouldn't go ahead >> and propose it if there's no interest in adopting such a thing in the >> long term. >> >> On Mon, Mar 10, 2014 at 11:19 AM, Emil Tin <e...@tin.dk> wrote: >> >> >> OSRM focuses on tags that are already in widespread use. From tag info: >> >> surface 8077811 >> tracktype 3212051 >> smoothness 208379 >> >> Even if a new tagging scheme is agreed on (by whom?) it would probably take >> quite a while before it's in common use worldwide. So for now I think the >> question is how OSRM should handle these 3 tags. >> >> >> >> On 10 Mar 2014, at 14:41 , Fernando Trebien <fernando.treb...@gmail.com> >> wrote: >> >> My personal point of view is: they mostly do, but in a needlessly >> complicated way. I think you'd be surprised at how far the discussion >> went (over 150 messages, many of which were quite long) to reach a >> simple agreement: deciding which tags/values to use in order to decide >> which roads are possibly in poor state, as to deserve special >> rendering. In this agreement, we settled on 3 tags (tracktype, >> smoothness and surface) to make such a decision. So it is clear that >> the community views the 3 tags as "necessary" for reasonable routing >> choices when reading the map visually. Trying to take any of the 3 out >> caused strong disagreement from certain people during that discussion. >> >> I tried to condensate my line of thought below, but it yielded a long >> text anyway. To encourage your reading, below is a link to the result >> at which I arrived after brainstorming. It establishes similarities >> with current tags and associating a subjective level of preference to >> each. This level was called "trafficability" during the other debate, >> but since then this name may be inadequate (it was used in a tag >> proposal). >> >> http://i.imgur.com/HUoE1iD.png >> >> In the beginning, I was almost convinced that the "surface" tag would >> be sufficient, but as other opinions came in, I was convinced that >> some of its values are too imprecise. "surface=unpaved", for instance, >> may refer to roads in excellent condition (specially if they'd be >> better described as "surface=compacted"), but also to roads likely in >> poor state (such as in "surface=dirt"). >> >> The Australian community seems to be recommending the use of >> "tracktype" for any road type besides highway=track which is what it >> was originally intended for, particularly within the German community. >> But then, many people use the "smoothness" tag for very similar >> reasons. It's easy to establish some rough correspondence between the >> two tags by reading the description of their values. It's easy to >> notice that smoothness provides more granularity at the "good" end of >> the spectrum (3 values representing the best conditions roughly >> correspond to a single value of tracktype) whereas tracktype has >> better precision at the other end (all of its other values correspond >> to a single value of smoothness). >> >> At the same time, the Australian community was trying to introduce new >> values for "tracktype" that correspond to other values of smoothness >> at the "bad" end of the spectrum. If these would not be accepted, they >> would pursue a new tag, "4wd_only=yes/no", that would correspond to >> those values and would be used for special rendering. Nobody seemed to >> be thinking of various transport modes, but some existing tags seemed >> to be doing this: mtb:scale for bikes, sac_scale for pedestrians, >> wheelchair for disabled people. >> >> So I thought: "if there was a single tag to represent all of this, >> would I be able to associate a level of preference to its values, with >> little doubt?" In other words, would the new classification system >> leave less, if any, doubts at all? Would it be sufficiently >> descriptive? It would in my experience, which includes: driving, >> cycling, walking and public transport in Brazil; driving, walking and >> public transport in North America; walking and public transport in >> Australia/NZ; cycling, walking and public transport in various places >> in Europe (England, France, Germany, Netherlands, Austria, Spain). I >> tested myself by associating such values comparatively, after having >> assigned each of the other tags a "class", producing the result I >> provided in the beginning. >> >> The question that I asked myself was: if I had to travel from A to B >> and there were two choices, a 100km-long perfectly flat asphalt road, >> and a shortcut with [surface characteristics here], how many km could >> this shortcut have at maximum to still look like a better choice? >> >> This measure would essentially mean a level of preference and directly >> translate into a coefficient multiplied to velocity in OSRM and other >> routers. Its inverse (1/value) would represent the level of effort. >> >> The obvious problem with this result: these values are my own opinion. >> For a public routing app (such as OSRM), one would have to sample more >> opinions, from people of different nations. But this is easier when >> you have a single tag than with various tag combinations. A single tag >> is also easier to teach and to map (which would encourage more people >> to describe the surface). And it solves well the rendering issues. It >> seems like a win for all involved sides: app developers, mappers, and >> users. >> >> Another little problem: only for class "5-grade2-pebblestone", I've >> forced the value up for thin-wheeled vehicles (bikes and wheelchair). >> The change was less than 10%, but still significant. I did this >> because I believed it would make more sense to have the preference >> curves asymptotically decrease for all vehicle types from one class to >> the next. (This actually suggests that thin-wheeled vehicles might >> require some slightly different classification system.) >> >> Of course I am open to suggestions on how these observations can be >> synthesized into a simpler tagging system. >> >> On Wed, Mar 5, 2014 at 5:17 AM, Emil Tin <z...@tmf.kk.dk> wrote: >> >> DO you mean a new osm tag? Doesn't the existing tags you mention cover >> surface quality? >> >> Med venlig hilsen >> >> Emil Tin >> IT- og Processpecialist >> Trafik >> _______________________________ >> KØBENHAVNS KOMMUNE >> Teknik- og Miljøforvaltningen >> Byens Anvendelse >> >> Njalsgade 13 Vær. 118 >> Postboks 380 >> 2300 København S >> >> Direkte 2369 5986 >> Mobil 2369 5986 >> Email z...@tmf.kk.dk >> EAN 5798009493149 >> -----Oprindelig meddelelse----- >> Fra: Fernando Trebien [mailto:fernando.treb...@gmail.com] >> Sendt: 28. februar 2014 17:35 >> Til: Emil Tin >> Cc: osrm-talk >> Emne: Re: [OSRM-talk] Beginner question: default car profile and >> tracktype/smoothness/surface >> >> Thank you Emil and Hans. I didn't know about the biking profile. Even though >> I'm a cyclist as well, I've been using the website mostly for car routing, >> and that's what OSRM is most known for here in Brazil. >> >> A while ago, I participated in a debate about making OSM-Carto use a >> different visual style to display roads in "worse than usually expected" >> state. As the debate developed, I made up a surface classification system >> that captures similarities among tags that represent "transit effort" >> (tracktype, smoothness, mtb:scale, sac_scale, wheelchair, 4wd_only, and >> surface) for various modes of transportation. I wonder if you'd be >> interested in something along this line, then I would go ahead and propose >> an official tag for it. >> >> On Fri, Feb 28, 2014 at 5:26 AM, Emil Tin <z...@tmf.kk.dk> wrote: >> >> >> >> Surface is already taken into account for bicycles in the OSRM main repo: >> >> https://github.com/DennisOSRM/Project-OSRM/blob/master/profiles/bicycl >> e.lua >> >> However, instead of multiplying, I found it more realistic to simply use the >> surface speed, instead of multiplying: >> >> surface_speeds = { >> ["asphalt"] = default_speed, >> ["cobblestone:flattened"] = 10, >> ["paving_stones"] = 10, >> ["compacted"] = 10, >> ["cobblestone"] = 6, >> ["unpaved"] = 6, >> ["fine_gravel"] = 6, >> ["gravel"] = 6, >> ["fine_gravel"] = 6, >> ["pebbelstone"] = 6, >> ["ground"] = 6, >> ["dirt"] = 6, >> ["earth"] = 6, >> ["grass"] = 6, >> ["mud"] = 3, >> ["sand"] = 3 >> } >> >> >> -- surfaces >> if surface then >> surface_speed = surface_speeds[surface] >> if surface_speed then >> if way.speed > 0 then >> way.speed = surface_speed >> end >> if way.backward_speed > 0 then >> way.backward_speed = surface_speed >> end >> end >> end >> >> Both approaches might have merit. >> >> >> >> Kind regards, >> >> Emil Tin >> IT- and Process Specialist >> Traffic Design >> ________________________________ >> CITY OF COPENHAGEN >> The Technical and Environmental Administration Traffic Department >> >> Islands Brygge 37 Vær. 118 >> Postboks 450 >> 2300 København S >> >> Telefon +45 2369 5986 >> Email z...@tmf.kk.dk >> EAN 5798009493149 >> >> >> -----Oprindelig meddelelse----- >> Fra: Hans Gregers Petersen [mailto:greg...@septima.dk] >> Sendt: 28. februar 2014 09:16 >> Til: osrm-talk@openstreetmap.org >> Emne: Re: [OSRM-talk] Beginner question: default car profile and >> tracktype/smoothness/surface >> >> Hi Fernando, >> >> I've always wondered if there are any plans taking surface >> type/quality into account in the default profiles. I live in a >> developing country (Brazil) with poorly maintained roads and these >> conditions make a big difference at the beginning and at the end of >> many routes if ignored. >> >> >> I do not know about the plans regarding the default profile, but I >> successfully used a simple "factor approach" to surfaces when doing our >> routing on bicycle paths here in Denmark. >> For instance setting the following in the LUA profile: >> >> -- How much does speed depreciate by surface surface_factors = { >> ["unpaved"] = 0.8, ["gravel"] = 0.8, ["cobblestone"] = 0.8, ["dirt"] = >> 0.8, ["earth"] = 0.8, ["sand"] = 0.8, ["cobblestone:flattened"] = 0.9, >> ["compacted"] = 0.9, ["fine_gravel"] = 0.9, ["wood"] = 0.9 } >> >> and then later adjuste the speed accordingly: >> >> -- Surface tag >> local surfacetag = way.tags:Find("surface") >> >> -- Surface factor >> if surface_factors[surfacetag] then >> way.speed = way.speed * surface_factors[surfacetag] way.backward_speed >> = way.backward_speed * surface_factors[surfacetag] end >> >> >> >> Best regards, >> >> Greg >> >> >> >> Hans Gregers Petersen >> Partner, Senior Consultant >> www.septima.dk >> >> _______________________________________________ >> OSRM-talk mailing list >> OSRM-talk@openstreetmap.org >> https://lists.openstreetmap.org/listinfo/osrm-talk >> >> _______________________________________________ >> OSRM-talk mailing list >> OSRM-talk@openstreetmap.org >> https://lists.openstreetmap.org/listinfo/osrm-talk >> >> >> >> >> -- >> Fernando Trebien >> +55 (51) 9962-5409 >> >> "The speed of computer chips doubles every 18 months." (Moore's law) >> "The speed of software halves every 18 months." (Gates' law) >> >> _______________________________________________ >> OSRM-talk mailing list >> OSRM-talk@openstreetmap.org >> https://lists.openstreetmap.org/listinfo/osrm-talk >> >> >> >> >> -- >> Fernando Trebien >> +55 (51) 9962-5409 >> >> "The speed of computer chips doubles every 18 months." (Moore's law) >> "The speed of software halves every 18 months." (Gates' law) >> >> _______________________________________________ >> OSRM-talk mailing list >> OSRM-talk@openstreetmap.org >> https://lists.openstreetmap.org/listinfo/osrm-talk >> >> >> >> _______________________________________________ >> OSRM-talk mailing list >> OSRM-talk@openstreetmap.org >> https://lists.openstreetmap.org/listinfo/osrm-talk >> >> >> >> >> -- >> Fernando Trebien >> +55 (51) 9962-5409 >> >> "The speed of computer chips doubles every 18 months." (Moore's law) >> "The speed of software halves every 18 months." (Gates' law) >> >> >> >> _______________________________________________ >> OSRM-talk mailing list >> OSRM-talk@openstreetmap.org >> https://lists.openstreetmap.org/listinfo/osrm-talk >> > > > > -- > Fernando Trebien > +55 (51) 9962-5409 > > "The speed of computer chips doubles every 18 months." (Moore's law) > "The speed of software halves every 18 months." (Gates' law) -- Fernando Trebien +55 (51) 9962-5409 "The speed of computer chips doubles every 18 months." (Moore's law) "The speed of software halves every 18 months." (Gates' law) _______________________________________________ OSRM-talk mailing list OSRM-talk@openstreetmap.org https://lists.openstreetmap.org/listinfo/osrm-talk