search

motional / nuplan-devkit

The devkit of the nuPlan dataset.
https://www.nuplan.org
Other
662 stars 126 forks source link

About the time out of _compute_trajectory #310

Closed yaqlee closed 1 year ago

yaqlee commented 1 year ago

When I ran docker compose locally, I logged the time out at 204th line in /nuplan_devkit/nuplan/planning/simulation/planner/remote_planner.py: function _compute_trajectory like this: Bo9yIzlQrt

the time out of the very first step is over 1s which caused failure in submission, and the time out of second step was reduced dramatically. Is there some difference between the first step and following steps?

jeVN3vDuTF

if the ckpt file was over 100M, would the time out of the first step increase?

piqiuni commented 1 year ago

Maybe you use any init in the first frame?

gianmarco-motional commented 1 year ago

@yaqlee, I'll take a look.

gianmarco-motional commented 1 year ago

@piqiuni is this the default sample scenario used when running docker compose?

In the first iteration you have no cache, so it is expected that it is slightly more expensive than subsequent samples. However I timed our baselines, and there should not be such a big difference between samples. Are you sure you are not initializing planner related attributes when computing the trajectory?