You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I believe for the nuPlan track the data is still provisional. However, I wanted to make sure to communicate some artefacts that I noticed when running alpasim evaluation locally on the complete set of scenarios on the navtest split.
Please refer to the video below. I see some cross traffic in the scenario which should not be the case if the aim is log replay simulation. Even if the simulation doesn't follow strict log replay, the correspoinding information (like bounding boxes) is missing in the BEV space which makes the evaluation error prone. Maybe you are aware of such issues but if you could also share how these scenarios will be resolved, that would be great!
I believe for the nuPlan track the data is still provisional. However, I wanted to make sure to communicate some artefacts that I noticed when running alpasim evaluation locally on the complete set of scenarios on the navtest split.
Please refer to the video below. I see some cross traffic in the scenario which should not be the case if the aim is log replay simulation. Even if the simulation doesn't follow strict log replay, the correspoinding information (like bounding boxes) is missing in the BEV space which makes the evaluation error prone. Maybe you are aware of such issues but if you could also share how these scenarios will be resolved, that would be great!
2021.05.25.14.16.10_veh-35_03373_03550-80a4b14aef3f5a52_f96d59bc-90cc-11f1-bac8-658e38d11a7a_CAM_F0_default.mp4