i understand turn penalties are not that far away, and what's essentially missing is just a good model. copenhagen university is intersted in helping out and provide parameters based on their research and data, but we need a better understanding of what such a model should consist of.

to get things started -
i imagine that there are 3 types of turns: normal turns, sharp turns, and u-turns, and that a penalty value is needed for each type of turn. so maybe something like:

side    type          angle range  penalty
--------------------------------------------
left    turn          20..160      0.1
left    sharp turn    140..160     0.3
left    u-turn        160..180     0.6
right   turn          -20..-160    0.1
right   sharp turn    -140..-160   0.3
right   u-turn        -160..--180  0.6

an alternative could be generalize this into a single 'continuous' formula that takes the angle as input and returns a penalty.

thoughts?