Hi,

Thanks for your great repo.

I try to run the code of TD3 on halfcheetah-v2 and I find the return is not very stable and below the result that you report.

More specially, I try to train the model with 1e6 steps (not 3e6 in the config file) and the return is about 9500.

So I wonder whether I need to tune the hyper-parameters.

Thank you very much. & Thanks for your contribution again.

Hi

I try to run the code of td3bc on antmaze-medium-play-v0 and I find the performance (is 0) is lower than the results in iql paper (10.6). May I ask the experimental setting about that?

Thank you very much!