Performance on sparse reward about dreamer HOT 2 CLOSED

Hi,
if I'm not mistaken, Dreamer has been applied on the "ball-in-cup" task from the DM control suite (for the task description, you can look here - just search for "cup"). I do not know, however if the episode is terminated on success, you would need to check that if you think it is important.
You can see the performance in the Dreamer paper - in my eyes it seems that it performs quite well.

Best,
André

from dreamer.

There are a lot more things that play into how well it works than weather the reward is only given at the last time step. Because there is a value function, it could work. The easiest way to find out is to run an experiment.

from dreamer.