Emergency braking at stop line with IDM and actionSteps

eclipse-sumo / sumo

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.

https://eclipse.dev/sumo

Eclipse Public License 2.0

2.55k stars 1.43k forks source link

Emergency braking at stop line with IDM and actionSteps #15443

Open kasop1 opened 2 months ago

kasop1 commented 2 months ago

test2.zip

Please see the attached file. The Krauss model does not have an emergency brake, but the IDM does. Why happened?

kasop1 commented 2 months ago

P.S. I also tried both EIDM and IDMM and in both cases the emergency braking occurred.

namdre commented 2 months ago

the simple fact is that, the whole code branch that computes safe accelerations when running with actionSteps isn't active for the IDM model.

namdre commented 2 months ago

Problem 1: the internal stepping is based on the step-length rather than action-step-length (i.e. step-length 0.1 and action-step-length 1 gives only a single internal iteration with default attribute stepping)

Problem 2: the internal stepping assumes constant speed rather than constant acceleration

Problem 3: the last speed from the internal stepping is used even though this may not be the best speed for the current situation

Domsall commented 1 month ago

@kasop1 Please use treaction when simulating with the EIDM. The documentation of all EIDM parameters can be found here.

@namdre Could you please add a warning, when actionStepLength is defined together with the EIDM? Thanks!

namdre commented 1 month ago

@Domsall

is the problem the same as explained for IDM (the iteration architecture?)
do you see this rather as a bug or as an inherent feature incompatibility for EIDM?

If the former, I'd rather put up a warning on the EIDM documentation page, if the latter, I'm fine with adding the warning in sumo.

Domsall commented 4 weeks ago

It is the latter: First, the antizipation of the EIDM does not output a constant acceleration until the next actionPoint. Secondly, the EIDM calculates a new "would"-acceleration next to the antizipated acceleration. The model can then update itself with treaction, even between actionPoints. See https://github.com/eclipse-sumo/sumo/issues/9681