Re: [sumo-user] Most realistic car following model?

[Hriday Sanghvi <sanghvih@xxxxxx>]

Hello again,

I have tried making the behaviour more 'rational' in SUMO by starting from scratch (vehicle parameters set to default) and only changing the following parameters.

First, I set lcMaxSpeedLatStanding=0 to disable the irregular behaviour.

Second, as advised previously, to prevent the vehicle from getting into such a situation, I set lcSpeedGainLookahead=100 (so it knows 100m in advance that it has to change lanes seeing the 'blockage' or 'stopped vehicle' for speed gain). Then there were times it could only partially complete the change, so I read that it may be because the change does not happen fast enough and the vehicle just gets stuck at the blockage.

Third, I set maxSpeedLat to a ridiculously high value to prevent partial changes. However, these changes did not seem to have the intended effect. I also played around with changing the minGap between vehicles (setting it to large amounts like 100), but that made the behaviour even worse. Maybe there is a way to tell vehicles to leave some space for other vehicles to change - and that is tied into the whole lcCooperative problem that I have previously asked about?

<vType id="veh" laneChangeModel="SL2015" lcSpeedGainLookahead="100" maxSpeedLat="999" lcMaxSpeedLatStanding="0" />

The flow is defined as:

<flow arrivalLane="current" arrivalPosLat="center" begin="0" departLane="random" departPosLat="center" departSpeed="random" id="f" number="100" route="r0" type="veh" vehsPerHour="10000" />

If I reduce the vehsPerHour from, 10000 to 1000, the vehicles are able to change lanes earlier to avoid the deadlock. But in the case of 10000 vehicles/hour demand, vehicles don't seem to get an opportunity to merge into the free lane and are often found 'trapped' and deadlocked at the blockage.

I am attaching the configuration files (.xml), the Python code (.txt), and a short video (.mp4) in a zip folder to show the problem. Please advise.

Thank you.

Sincerely,

Hriday

On Thu, 7 Apr 2022 at 14:55, Jakob Erdmann <namdre.sumo@xxxxxxxxx> wrote:

The proper solution is outlined in the updated comments of #3392 but it actually depends on fixing at least one other problem first so it may take some time.

Am Di., 5. Apr. 2022 um 09:36 Uhr schrieb Hriday Sanghvi via sumo-user <sumo-user@xxxxxxxxxxx>:

Hello Jakob, thank you for your response.

I tried setting lcMacSpeedLatStanding=0, and I am indeed facing a deadlock as mentioned In this issue: https://github.com/eclipse/sumo/issues/3392#issuecomment-902466679 . Is there a way to fix the deadlock issue without manually messing about the lane-changing parameters? Maybe some known set of parameters that generically work for all situations without causing a deadlock while also preventing sliding - I also tried increasing the lcSpeedGainLookahead to no avail!

Sincerely,

Hriday

On Tue, 5 Apr 2022 at 06:43, Jakob Erdmann <namdre.sumo@xxxxxxxxx> wrote:

> the SL2015 LC model combined with Krauss CF model makes for some unrealistic lane changing when faced with a blockage in a 2-lane condition (slides laterally to lanes when there's no space to accelerate in front).

This is a general issue of SL2015 and not limited to Krauss. You can may be able to prevent sliding by setting lcMacSpeedLatStanding=0

> Both ACC and CACC seem to be more realistic and intelligent. Are these modeled close to the state-of-the-art systems that self-driving vehicles use currently for cruise control?

I'd say no. The models were developed/tested for a rather narrow range of applications and are known to behave badly (i.e. collide) outside that range. To learn more, please refer to the publications linked in their documentation.

I you need a model that is more sophisticated than Krauss, take a look at https://sumo.dlr.de/userdoc/Car-Following-Models/EIDM.html

regards,

Jakob

Am Di., 5. Apr. 2022 um 01:15 Uhr schrieb Hriday Sanghvi via sumo-user <sumo-user@xxxxxxxxxxx>:

Hello,

I have noticed that the Krauss CF model seems to outperform other models like IDM, ACC and CACC in terms of total simulation time for all vehicles to reach the end of their journey. But the SL2015 LC model combined with Krauss CF model makes for some unrealistic lane changing when faced with a blockage in a 2-lane condition (slides laterally to lanes when there's no space to accelerate in front).

Both ACC and CACC seem to be more realistic and intelligent. Are these modelled close to the state-of-the-art systems that self-driving vehicles use currently for cruise control?

Thank you.

Sincerely,

Hriday.

_______________________________________________
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user

_______________________________________________
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user

_______________________________________________
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user

_______________________________________________
sumo-user mailing list
sumo-user@xxxxxxxxxxx
To unsubscribe from this list, visit https://www.eclipse.org/mailman/listinfo/sumo-user

Attachment: lc_behaviour.zip
Description: Zip archive

Attachment: 138.png
Description: PNG image