search

Xingyu-Lin / mbpo_pytorch

A pytorch reprelication of the model-based reinforcement learning algorithm MBPO
150 stars 38 forks source link

cannot reproduce Again #10

Open betray12138 opened 1 year ago

betray12138 commented 1 year ago

image As the upper figure shows, the walker2D environment's performance can not match the figure you put in the README. What could I do ?

Sen1553 commented 10 months ago

Hello. Do you tackle it? I met the same problem. Could you tell me the solution.

betray12138 commented 10 months ago

Hello. Do you tackle it? I met the same problem. Could you tell me the solution.

No, I don't. However, I found that the torch version proposed by mbrl repo is more stable, so I choose to use mbrl-lib

Sen1553 commented 10 months ago

Hello. Do you tackle it? I met the same problem. Could you tell me the solution.

No, I don't. However, I found that the torch version proposed by mbrl repo is more stable, so I choose to use mbrl-lib

Thanks for your suggestion