Eclipse SUMO is an open source, highly portable, microscopic and continuous traffic simulation package designed to handle large networks. It allows for intermodal simulation including pedestrians and comes with a large set of tools for scenario creation.
Currently, simulation routing only reacts to (averaged) observed vehicle speeds.
This means, vehicles only change their route once the traffic situation has sufficiently degraded. The "signal" to avoid an edge comes late and some congestion occurs (though the route choices at least prevents it from becoming worse).
As a consequence, the resulting assignment may not be as robust as desired. Since some jamming occurs, small changes in vehicle behavior can easily create worse jamming when replaying the recorded routes: #3024.
A more robust routing would avoid edges near congestion even while they are still in a free flow state.
Detecting a close-to-congested state might be possible by comparing observed speed in relation to free-flow speeds. However, this would still be a "late" signal since there is a large range of traffic states with free flow speed. (A simple idea for pessimism would be to magnify the difference between free flow speed and observed speed by a configurable factor).
Alternatively, impeding congestion could be detected by traffic volume rather than by speed.
For this we need a mapping from volume/density to speeds.
Such a mapping would offer a pathway for adding "pessimism" (robustness) by using the speed associated with volumes that are bit larger than the currently observed volumes (or by applying a corresponding reduction to the currently observed speed).
This could be interpreted as a forethought along the lines of "this road is already quite busy. With a bit more traffic it will be congested, better not drive there".
A mapping from densities to speeds (for each edge) could either be obtained from
a prior simulation run (edgeData)
an analytic function (fundamental diagram)
an online calibration - each simulation step offers new data points (some congestion might be seen but afterwards the system would know to avoid it)
namdre
commented
2 years ago
Currently, simulation routing only reacts to (averaged) observed vehicle speeds. This means, vehicles only change their route once the traffic situation has sufficiently degraded. The "signal" to avoid an edge comes late and some congestion occurs (though the route choices at least prevents it from becoming worse). As a consequence, the resulting assignment may not be as robust as desired. Since some jamming occurs, small changes in vehicle behavior can easily create worse jamming when replaying the recorded routes: #3024.
A more robust routing would avoid edges near congestion even while they are still in a free flow state.
Detecting a close-to-congested state might be possible by comparing observed speed in relation to free-flow speeds. However, this would still be a "late" signal since there is a large range of traffic states with free flow speed. (A simple idea for pessimism would be to magnify the difference between free flow speed and observed speed by a configurable factor).
Alternatively, impeding congestion could be detected by traffic volume rather than by speed. For this we need a mapping from volume/density to speeds. Such a mapping would offer a pathway for adding "pessimism" (robustness) by using the speed associated with volumes that are bit larger than the currently observed volumes (or by applying a corresponding reduction to the currently observed speed).
This could be interpreted as a forethought along the lines of "this road is already quite busy. With a bit more traffic it will be congested, better not drive there".
A mapping from densities to speeds (for each edge) could either be obtained from
Related Issues: #4635, #6746, #9043