Request for official visualization & evaluation code to access raw trajectory, input data besides Driving Score JSON

[chan-yuu]

Hi,

Thanks a lot for releasing the great CaRL codebase and the corresponding paper.

Currently after running evaluation, I can only get a single JSON file containing the final Driving Score metric. However, I cannot obtain any intermediate data needed for visualization and detailed analysis mentioned in the paper. For example:

• Raw vehicle trajectory of each evaluation episode
• Original model input data (observations, sensor inputs, planning context, etc.)
• Per-step reward values, violation records and intermediate planning outputs

I wonder if the official repository provides scripts or functions to export these complete evaluation data for plotting trajectory figures and reproducing the visual comparisons shown in the paper?

Looking forward to your reply, thank you very much!

[Kait0]

The code for the visualizations in the paper is included.

See here on how to turn on debug outputs.

[chan-yuu]

oh I see. Thank you.