Largely overestimated travel times in routing for single vehicles departing from standstill at simulation start

[m-kro]

Discovered in Huainan network. When a single vehicle departs from standstill short after simulation start on an edge and reroutes immediately, it is considered in the edge mean speed computation. Thus: v=0 is remembered in MSRoutingEngine::myEdgeSpeeds and the edge travel time becomes edge_travel_time = edge_length / 0.001 due to

sumo/src/microsim/devices/MSRoutingEngine.cpp

Line 164 in db9551c

return MAX2(e->getLength() / MAX2(myEdgeSpeeds[id], NUMERICAL_EPS), e->getMinimumTravelTime(v));

[namdre]

there is no easy fix:

• initializing with speed limit instead of meanSpeed makes the simulation very unresponsive to early jamming
• initializing with speed limit and increasing initial responsiveness (by initially scaling with the number of observed weight-updates rather than the steady-state number of updates) increases responsiveness at the cost of responding to "noise"
  diff.14835.txt