Models that have an explicit goal state, and a distance estimator, are naturally going to be more efficient than vanilla RL without this side information where

Models that have an explicit goal state, and a distance estimator, are naturally going to be more efficient than vanilla RL without this side information where it has to learn purely by exploration. Edit: Im saying it looks like apples vs oranges.