Question about training tasks used.

[zlm646]

Hi:
Thanks for your excellent work. I justed noticed in NavDP training code that you only used no goal navigation task along with point goal navigation task, not like you mentioned in the paper that using all four tasks for training achieves the best result, can you explain the reason behind this mismatch please?

[wzcai99]

Hi, currently, the provided training code provides a minimal training validation, and without the other two losses, you can still get a satisfying policy performance. We will update the training code with the other two losses recently. Please stay tuned.

[zlm646]

Got it, thanks for replying.