Clarification on GOAT Benchmarking against your method UniGoal

[sktalha68]

Hello,

Thanks for releasing the code for this work. I had a question regarding the benchmarking of GOAT reported in the paper "UniGoal: Unifying Long-Horizon Navigation Tasks in Embodied AI".

In the paper, you report results for GOAT. Could you clarify:

1. Did you run GOAT’s experiments yourselves for benchmarking?

2. If yes, where did you obtain the GOAT model and weights? Did you train it yourself or used a pretrained checkpoint? Can you please point me in the right direction for reproducing those numbers?

3. Were there any modifications made to the original GOAT implementation for these experiments?

This information would be very helpful for reproducing the results. Thanks for your time and for sharing your work.