rlworkgroup / garage

A toolkit for reproducible reinforcement learning research.
MIT License
1.84k stars 309 forks source link

maml_trpo not good as trpo in PointEnv environment? #2298

Open tianyma opened 2 years ago

tianyma commented 2 years ago

Hi there, I am currently on the master branch, I test maml_trpo and trpo on garage env PointEnv to see if maml_trpo works better than trpo, but I found the result not good as expected. I post the average return and success rate below. Here are maml results.

20210730_09:40:26_001 20210730_09:43:32_001

Here are trpo results. 20210730_09:53:30_001 20210730_09:53:53_001

this is quite confusing, cause I thought the success rate of maml reaches 1.0 faster than trpo.