Request for per-episode evaluation results on VLN-CE

[Abed-Luo]

Hi, thank you for releasing this excellent work!

I am interested in conducting a more detailed analysis of Qwen-RobotNav's performance on the VLN-CE benchmark. Would it be possible to provide per-episode evaluation outputs, in addition to the aggregated metrics?

For example, the following information would be very helpful:

• Episode ID
• Predicted trajectory or visited viewpoints
• Final stopping position
• Success / Oracle Success
• Navigation Error
• SPL
• Model-generated actions or responses at each step

Thank you very much for your time and for sharing this work!