Thank you for this excellent work! I have a few questions I hope the authors can clarify.

In the “Robotic Navigation” subsection on page 10 of the arXiv preprint, NVILA is proposed as a new baseline that can be fine-tuned.

1. Could LoRA (or similar parameter-efficient methods) be used for fine-tuning?
2. Would you be willing to share the training script to help the community reproduce and build on your results?
3. Is the dataset the same one employed in your prior work, NaVILA?

Looking forward to your reply.