The problem of stop_infraction in Carla simulation

[Forgotten0116]

Hi! first of all, thank you for your great work, and I also really appreciate your earlier replies about CARLA-related code.
I’m a beginner in end-to-end autonomous driving research. Recently I managed to run the LAW model successfully in CARLA. However, after training, I consistently observe a large impact from "stop_infraction" in the CARLA evaluation. This happens for both LAW and TCP in my setup, and I’m not sure how to address it.

Here is a snippet of the aggregated evaluation infractions I’m seeing:
"infractions": {
"collisions_layout": 0.0,
"collisions_pedestrian": 0.0,
"collisions_vehicle": 0.06063314561294968,
"outside_route_lanes": 0.010083096735781095,
"red_light": 0.009821689108545134,
"route_dev": 0.0,
"route_timeout": 0.0,
"stop_infraction": 0.11556071665065652,
"vehicle_blocked": 0.0
},

My CARLA version is 0.9.10.1 (same with TCP) and trainning with the dataset provided with TCP.
Do you have any recommended practices to address the problem? Any guidance or pointers would be greatly appreciated.
Thanks again for your time and for sharing your work.

Best regards