Packed CARLA 0.9.15 benchmark for autonomous-navigation continual learning under domain shift.
Simulator: CARLA 0.9.15
Towns: 7
Compact domains per town: 7
Expected base tasks: 49
Expected episodes: 490
Expected frames: 490,000
Format: tar shards tracked with Git LFS
Episode ordering: domain_major
Modalities: RGB, semantic segmentation, LiDAR, vehicle state
0 clear__NPC00_none
1 night__NPC00_none
2… See the full description on the dataset page:
https://huggingface.co/datasets/checo1092/Continual_Learning_for_Autonomous_Driving.
