Datasets:
Languages:
English
Size:
100K<n<1M
Tags:
autonomous-navigation
autonomous-driving
imitation-learning
continual-learning
domain-shift
semantic-segmentation
DOI:
License:
File size: 1,382 Bytes
e49b848 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 | # Dataset illustrations
## Weather and background traffic

Two Town02 starting slots: spawn 84 (training, top) and spawn 5 (test, bottom). Columns: clear and hard rain without traffic, followed by clear and hard rain with vehicles. Episodes with traffic requested 50 vehicles. Frames retain the full field of view; poses within each row differ by at most 0.57 m and 0.6 degrees. See [frame identifiers](assets/weather_traffic_frames.json) and [PDF](assets/weather_traffic.pdf).
## Trajectories by split

Seven towns, two domains (clear without NPCs and clear with vehicles and pedestrians), ten starting slots per town/domain: 140 episodes with 1,000 saved positions each. Colours identify train/validation/test assignments; dots identify starts. Both conditions within a town share a spatial scale; scale bars differ between towns. Nearby paths may occupy different lanes or directions. Distinct starting slots do not imply disjoint routes. [PDF](assets/trajectories_by_split.pdf).
## Reproduction and CLIP characterization
[Figure-generation scripts and CLIP diagram](https://github.com/checo1092/egospatial-cl-code/blob/main/docs/publication_figures.md) are provided in the code repository. Images in this dataset repository are covered by its CC BY 4.0 license.
|