Fixing unroutable locations

[leonardehrenfried]

In Herrenberg and the surrounding regions there are a number of locations that are unroutable for cars:

URL prod	URL dev	✔️
URL prod	URL dev	✔️
URL prod	URL dev	❌

It’s not entirely clear why that is but a frequent source for unroutable locations are “islands”: small clusters of roads that OTP cannot route to or out of because they are not connected to the road network, while being reachable by bike or for pedestrians. A cause for this can be incorrectly mapped roads, gates or bollards.
In any case, whilst it isn’t clear that the above locations contain mapping errors, even if they do, it should not lead to unroutable locations but rather to sub-optimal routing results.

Related to #46

[leonardehrenfried]

Helsinki reported that they have a similar problem and made a PR for OTP1: opentripplanner#2676

They are planning to port it to OTP2 but cannot really say when that happens.

[flaktack]

The routing issues seems to be caused by three things:

• linking to non-driveable edges (Drive only requests are linked incorrectly opentripplanner/OpenTripPlanner#3428)
• thru-traffic areas within no-thru traffic areas (should be fixed by Island pruning opentripplanner/OpenTripPlanner#3426)
• handling of thru-traffic areas needs to be expanded to allow both starting/ending in no-thru traffic areas, but not entering and leaving them (issue / PR in the works)

[flaktack]

I've pushed a branch with all the related fixes: feature/routing-test, which contains the drive-only linking fixes, no-thru traffic render and island pruning for car only (not bicycle).

These are the linked trip-plans using the debug client:

• ✔️ trip-plan
• ❌ trip-plan - this only works in one direction,
• ✔️ trip -plan

This trip-plan also only works in one direction -- we working on creating explicit test to see why the two tests fail in one direction only, and fixing it if the reason is clear.

[leonardehrenfried]

@hbruch What is the expected routing result when you want to drive to Marktplatz? As far as I can see its surrounded by either pedestrian areas or no-traffic areas.

Should the result just snap to the nearest drivable road?

[hbruch]

Should be park and walk

[leonardehrenfried]

Just to clarify: in the car tile in DT you want to see the car being parked in the street and then walk? I don't think this is how it works right now. The driving leg is always the latest one.

[leonardehrenfried]

Or do you mean something like this?

[hbruch]

In the car tile, but neverthless parked in a garage and the walk. However, I wonder if there isn‘t a better parking than the one at the station, eg Bronntor?

[leonardehrenfried]

There is but the ParkAPI updater is not complete yet: #55

[leonardehrenfried]

Sorry to belabour the point, but you want Park & Ride results in the car tile?

This is currently not possible as the mode is CAR and not CAR_PARK.

[hbruch]

I‘m aware, that technically car_park returns park & walk and park & transit. But from a user perspective, car and walk is not park & ride.

As a user, I want to see car & walk in the car tile, car & transit in the park & ride tile

[leonardehrenfried]

Not sure I can immediately come up with a good implementation strategy. Can you?

[flaktack]

For the CAR tile depending on whether to is inside a no-car-zone (some polygon) one of two request may be sent: CAR_PARK or CAR (without any TRANSIT modes).

For the P+R tile CAR_PARK,TRANSIT (any TRANSIT type works, there just has to be something) could be modified to not return directMode = CAR_TO_PARK results:

OpenTripPlanner/src/main/java/org/opentripplanner/api/parameter/QualifiedModeSet.java

Lines 145 to 149 in d2687d9

	} else if (requestMode.qualifiers.contains(Qualifier.PARK)) {
	accessMode = StreetMode.CAR_TO_PARK;
	transferMode = StreetMode.WALK;
	egressMode = StreetMode.WALK;
	directMode = StreetMode.CAR_TO_PARK;

directMode = transitModes.isEmpty() ? StreetMode.CAR_TO_PARK : null

Would this work?

[leonardehrenfried]

Estimation: 0/2/1 days