tristandeleu/pytorch-maml-rl

Question about the Ant env

Opened this issue · 2 comments

jzstudent commented

Hi, thanks for your code which gives me a lot of help.
But I have a question about the AntVel env, when I tried to train MAML in this environment, I can't get a good result. But it works well in the HalfCheetahVel enviroment. Did you encountered this problem？
And could you tell me how to deal with it? Thanks!

tristandeleu commented

Hi! Yes that's my experience as well, I haven't been able to reliably get good results on AntVel with this implementation, even though it seems to work well on AntGoal and HalfCheetahVel. That was already the case prior to the refactoring of this repo earlier this year. The environment is almost identical to the original maml_rl repo, the only difference being that the environments here are based on Gym environments, as opposed to Rllab environments (but if I remember correctly both are identical). I tried the Rllab environments directly in #16, but I didn't go very far with that.

I don't have any good solution unfortunately. You could try the Rllab environments from #16, see if that makes any difference. Let me know if you find anything!

jzstudent commented

Thanks very much! I will try the Rllab env and see how it turns out! :)