search

facebookresearch / deep_bisim4control

Learning Invariant Representations for Reinforcement Learning without Reconstruction
Other
144 stars 37 forks source link

Question about augmentation #10

Closed tunglm2203 closed 3 years ago

tunglm2203 commented 3 years ago

Hi. Thanks for sharing your work.

Could I ask did you use augmentation in your method? Since I didn't find any implemented transformation in your implementation. But in the appendix, you mentioned:

We modify the Soft Actor-Critic PyTorch implementation by Yarats and Kostrikov [33] and augment with a shared encoder between the actor and critic.

Am I understanding wrong?

Thank you in advance.

amyzhang commented 3 years ago

Hi!

We do not use data augmentation in our method. The word augment was used here to mean that we modified the original implementation with a convolutional encoder that is shared between the actor and critic, to adapt SAC to the pixel setting.