About Scenarios in training data

[autonomousvision]

About Scenarios in training data

[ap229997]

That's a good question. While experimenting with CARLA 0.9.10, I had observed that the events are triggered by background agents, not the agents defined in 'other_actors'. I was also surprised by this and a nice way you could verify this is to spawn a single scenario (without any 'other_actors' but make sure there are background agents) yourself. You should be able to see the scenarios spawning correctly. Another interesting thing to do would be to remove any background agents but add a single agent in 'other_actors'. I suspect that in this case 'other_actors' would be used to spawn the scenario. I had not tested this specific case so you could try to verify this too. I don't know if the behavior of the scenarios has changed in the newer CARLA versions.

[penghao-wu]

Oh that is surprising since I think the background agents are spawned randomly

[ap229997]

I think it's not exactly the background agents but the agents are explicitly specified in the definition of the scenarios, e.g. check follow_leading_vehicle.py

[penghao-wu]

Oh I see. Thank you very much! Congratulations on your good places with transfuser based methods on Carla challenge :)