Experiment setting of Fig6 in FAST paper

[yfpeng1234]

Dear author,
Thanks for open-source this inspiring work. I'm trying to reproduce the result in Fig 6 Libero of "FAST" paper. Specifically, I hope to know the following experiment settings in Libero:

1. the observation window (how many history obs are used)
2. the action prediction window, is it the same with the default setting in this paper?
3. the actual action execution window
4. are the training steps the same for Naive, FSQ, FAST and FAST+? or different training steps to make sure they all converge
   Thanks a lot!

[kpertsch]

Hi!

1. no history, just the current step
2. I believe we predict action chunks of 10 steps and execute all 10 steps
3. see above
4. all use the same, 40k iterations @ 256 batch size I believe

Closing for now since i hope it answers all questions here!