Possibility of using in RL setup

[ugurkanates]

Hello , very nice work congrats.
I was wondering did you guys tried this (or anyone reading this issue) as pre trained weights for say off policy algorithms such as QT-Opt. I was wondering would it be possible to use this and with some fine tuning (RL) to get adapt to other real time situations with grasping that are not exclusive to your case

https://sites.google.com/view/efficient-ft/home

They used imagenet(alex net) as baseline but would it be possible to replace your network here?