Support Left-Sided Driving

[MoKob]

This requires OSRM to be country aware.

At the moment, we only support right-handed driving per default and assume it to be the general mode of driving.
This can result in broken instruction handling in left-handed driving countries.

When we get some sort of country-awareness into OSRM, we need to have a look at guidance and make sure to address left-handed driving.

[danpat]

@MoKob Can you list some specific examples?

[MoKob]

https://github.com/Project-OSRM/osrm-backend/blob/develop/src/extractor/guidance/roundabout_handler.cpp#L74 would be such a case.

[MoKob]

In general, when I noticed it during coding, I labeled the locations with //FIXME items

[willwhite]

In the case of the roundabout handler above, what would be broken? For example, would the guidance just say "turn right out of the traffic circle" instead of "turn left out of the traffic circle"?

[MoKob]

@willwhite the count for intersections is wrong in some cases. it may say take the first instead of the second. The angles themselves are independent of the side of driving.

[emiltin]

related: #333

[miccolis]

@MoKob any sense of what follow up work will remain after #4415 merges?

[MoKob]

I am not aware of many situations that actually change in left-sided conditions. We already have some roundabout-handling based on profile flags. This would need to be implemented on a byCountryFlag basis, rather than on global level. The main handling works, we just need to ensure the behaviour is adjusted based on countries, so we can have a single graph with both left/right handed driving.

Turn lane handling for roundabouts assumes right-sided driving. We currently have them disabled anyhow, though. When we redo them to work at all, we should take left-sided driving into account.
Other cases might be optimised for right-hand detection. The code requires a general overhaul though, which could be best done, after we have maneuvers in extraction, since lane handling on OSRMs data model is a game of chance right now.

With this flag, we would also be able to switch to a TurnType::UTurn and DirectionModifier::Left\Right. The main issue here is, that this would constitute a breaking API change, since it cannot be interpreted as a normal turn. A former uturn might suddenly be announced as a Turn Left.

[daniel-j-h]

Lane Anticipation:

osrm-backend/src/engine/guidance/lane_processing.cpp

Line 193 in 11e7b6e

else // FIXME: right-sided driving

and I came across a few comments sprinkled throughout the code base a while ago

osrm-backend/src/extractor/guidance/intersection_normalizer.cpp

Lines 259 to 260 in 288155e

	// FIXME if we have a left-sided country, we need to switch this off and enable it
	// below

osrm-backend/src/extractor/guidance/turn_lane_handler.cpp

Line 279 in 5ece65c

// FIXME the lane to add depends on the side of driving/u-turn rules in the country

[cmhh]

This is a massive issue for left-side driving countries (of which there are a lot!). I'm trying to use the trip service to solve the TSP, and the results are unusable--the sorted order of the waypoints is essentially in the reverse order from what would be correct in most cases. I have set approach to curb for all waypoints, but the result has them all visited on the right-hand side of the road. I could be missing something, of course?...

Perhaps approach could take left, right, or unrestricted; rather than curb, or unrestricted?

[danpat]

@cmhh OSRM now has basic support driving side. The demoserver calculates driving side during pre-processing and returns it with routing results.

If you're running your own OSRM server, you'll need to add the --location-dependent-data data/driving_side.geojson flag to osrm-extract so that it performs the lookup for each road.

As noted above, there are probably still some corner cases around guidance, but the approach=curb property should work.

If you're seeing a specific problem, please open a new ticket with a recipe to reproduce the data you're using, a sample query, the results you're seeing, and the results you expect.

[cmhh]

Great! I was looking down the barrel at doing this with ArcMap's Network Analyst (yuck), or implementing something bespoke myself. I made a Docker image with this flag set and it seems to be working pretty well.

Thanks heaps.

[murphy-paul]

I've been trying to enable the above mention support for driving side on our OSRM service using the location-dependent-data but the response is still returned with driving_size = right. I've also tried this on the OSRM demo server and the result is the same. For coordinates in Ireland the driving side should be left

curl --location --request GET 'http://router.project-osrm.org/trip/v1/driving/-7.700471878051758,52.35865823116153;-7.700697183609009,52.3519350813739?overview=full&steps=true&roundtrip=false&source=first&destination=last'

I'm using docker to process the osrm files and run the routing engine. Commands I use are below. II'm using the latest docker builder (image id: d45a54a4bc52)

docker run --rm -t -v "${PWD}/data:/data" osrm/osrm-backend osrm-extract --location-dependent-data "/data/driving_side.geojson" -p "/opt/car.lua" "/data/britain-and-ireland-latest.osm.pbf"
docker run --rm -t -v "${PWD}/data:/data" osrm/osrm-backend osrm-partition "/data/britain-and-ireland-latest.osrm"
docker run --rm -t -v "${PWD}/data:/data" osrm/osrm-backend osrm-customize "/data/britain-and-ireland-latest.osrm"
docker run -t -i -p 5000:5000 -v "${PWD}/data:/data" --name osrm_test osrm/osrm-backend osrm-routed --max-trip-size 1000 --algorithm mld "/data/britain-and-ireland-latest.osrm"

The only way I could get driving_side to be left is to change the left_hand_driving of the profile to be true.

Any help would be greatly appreciated.

[danpat]

Looks like driving_side.geojson as messed up in commit 3c607a4 - there should be "properties": { "driving_side": "left" }, on each feature in that .geojson file.

Try adding "driving_side": "left" to the properties of each geojson feature in that file and see if it fixes your problem.

[murphy-paul]

That's it. That fixed the problem for me. Thanks again for the super fast response @danpat .

[danpat]

If you have time, a PR fixing the file in the repo would be appreciated!

[murphy-paul]

I don't have permissions is to push a branch. Do I need to do this on a fork @danpat ?