Implement living street avoidance

[SvetlanaBayarovich]

Issue

Introduces use_living_streets parameter which allows to tune living streets preferences during routing with different costings.
The impact on route selection consists of 2 parts:

• penalize entering living street areas, if parameter value is in range [0, 0.5]
• increase or decrease costs for living street edges according to parameter value.

Tasklist

• Add tests
• Add #fixes with the issue number that this PR addresses
• Update the docs with any new request parameters or changes to behavior described
• Update the changelog

Requirements / Relations

Link any requirements here. Other pull requests this PR is based on?

[SvetlanaBayarovich]

Rue de Thorigny is a living street, so avoid it now

[kevinkreiser]

but this is a bike share route which means its just pedestirans and bike riders, we shouldnt be avoiding living streets for those modes of travel should we?

[SvetlanaBayarovich]

Yeah, but this is the test case for auto costing, as the line 270 says, so I thought this was expected

[SvetlanaBayarovich]

Some test cases from initial issue:

prod	result	data

[purew]

This test is great, easy to read and follow.

[SvetlanaBayarovich]

All credit goes to @genadz and his tests on track avoidance :)

[purew]

Nice, LGTM

[kevinkreiser]

should this be 1.f? isnt the point of most of these living streets that pedestrians get right of way at or above other modes of travel?

[SvetlanaBayarovich]

I agree with your idea, but in my opinion pedestrians are more sensitive to route distance/duration increase which can happen if we favor living streets more. That's why I didn't want to mess with this costing a lot

[purew]

Interesting, so a customer who would like more scenic and clean routes may want to crank this up at the potential expense of time and distance. Makes sense.

[SvetlanaBayarovich]

According to the testing, the amount of routes that changed significantly is relatively small. There are only 62 routes with relative difference in distance greater than 1%. As for the weight, the routes with high difference are all of a certain type: short (from 300-400 meters to couple of kilometers) and with mandatory living street edges on them, so that the route itself does not change even change wrt master, only the cost increases. Thus when living streets are not mandatory, we manage to avoid them without significant weight increase.

The most interesting cases I came across are

1)Well, we do avoid this living street. The result just does not look nice.

master	living street avoidance	data

2)More relevant for the large areas of living streets. We add sth like 400 meters to the total distance to save like ~50 meters of riding living streets (because 'living street' edges are really costly)

master	living street avoidance	data

[SvetlanaBayarovich]

Testing another approach with high penalty for transition into living street and not such a big factor (previously I was betting everything on large factor) in order to avoid situation 2) from my comment above. The idea is to minimize the part of the route that goes through living streets, but not to add to much to the route length if we need to use it as an origin or destination