consider occupancy when computing bestlanes

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

2.58k stars 1.44k forks source link

consider occupancy when computing bestlanes #7157

Open namdre opened 4 years ago

namdre commented 4 years ago

as a tie-braker for lanes that are otherwise equally good.

This would avoid issues where a lane has multiple follower lanes on the same edge but lane-changing is not possible (e.g. because they are separated by prohibited lanes or we finally implemented #3656)

related: #4108

namdre commented 4 years ago

occupancy is temporary whereas bestlanes may be kept for longer (i.e. until leaving an edge). Still, this might be better than using the current computation with arbitrary choice.

namdre commented 1 year ago

see also #13392

namdre commented 1 year ago

bias in formation of queues due to laneChange prohibitions observed in #13300

namdre commented 7 months ago

Typically, bestLanes are computed when entering a new lane. At this time the occupancy is not very predictive of the downstream occupancy when leaving that lane.

namdre commented 7 months ago

Result of discussion with @behrisch:

when computing bestLanes, record the information that a tacticalDecisionPoint was encountered after a lane (two or more strategically equivalent continuations)
during planMove, if a car is forced to slow down due to a leader on a subsequent lane and a tacticalDecisionPoint lies upstream of that leader, recomputeBestLanes with additional considerations for the tactical situation (afterwards delete the tacticalDecisionPoint to avoid repeated recomputation).

namdre commented 1 month ago

Still, this might be better than using the current computation with arbitrary choice.

At the moment, the choice isn't even arbitrary but instead fixed to pick the rightmost lane among all equally good candidates. Even picking a random lane could be better.