The MineRL environment is obviously very different from the demonstration environment. For example, the humans that provided the demonstrations had a higher-resolution view, had audio input, could see and restore health, could craft ladders and swords, etc. Of course, transfer learning may be valuable despite these differences.
One difference that seems significant and may be worth fixing is that MineRL agents are not able to put torches on walls. It’s funny to see. Do y’all have any plans to fix that bug with action['place'] = 'torch'?
And, a secondary question: Does action['nearbySmelt'] = 'coal' actually do anything?
Fair points. The task is technically doable with just given items, but indeed it is difficult to navigate underground based on visual information when there is nothing visible :). Like you pointed out, the fact that demonstrations can have these makes it worse. I can see the dilemma though: including all of the different actions would make the training a lot harder (more confusing actions that have little relevance towards obtaining diamond), and the training is hard-enough as it is.
To be clear, I’m not asking for any actions to be added. Fixing the broken ones would be appreciated though.
Thanks for the link about smelting coal! I like the theory that the action creates charcoal, which may not show up in the inventory. I’ll look to see if the log/plank counts decrease when the action is taken.