jcwleo/curiosity-driven-exploration-pytorch

Small internal rewards

Opened this issue · 2 comments

postmachines commented

Int reward values are anomaly small for sparse reward environments.
This is normal? if so, why?

Venture example:
Снимок экрана 2019-04-23 в 1 05 35

JACKYWANG12345 commented

Hello, I also encountered the same problem as you. How did you solve it in the end? The rewards from training are the same as yours. I hope to get your reply, which will be very helpful to me.

postmachines commented

Hello, unfortunately I don't remember. It was more then 3 years ago:( Try another repo, I guess there are many new curiosity driven agents now