robodhruv / visualnav-transformer

Official code and checkpoint release for mobile robot foundation models: GNM, ViNT, and NoMaD.
http://general-navigation-models.github.io
MIT License
425 stars 56 forks source link

Question about yaw learning #31

Closed evgeniishch closed 1 day ago

evgeniishch commented 1 month ago

Good afternoon!

I was curious, why did you choose to include yaw change ψ to the intermediate action space in ViNT? Does it bring extra quality compared to just predicting nomalized actions xy? It's interesting that you do not predict angles in nomad. What was the motivation for that?

Thank you for great papers anyway ^_^

ajaysridhar0 commented 1 day ago

Glad you enjoyed the papers! We added yaw predictions to ViNT as an extra supervision signal. However, we found no performance difference and removed it for NoMaD.