search

toshikwa / gail-airl-ppo.pytorch

PyTorch implementation of GAIL and AIRL based on PPO.
MIT License
189 stars 30 forks source link

The results are not good on other Mujoco environments #2

Closed guanweifan closed 2 years ago

guanweifan commented 2 years ago

Hi, thanks for your wonderful project! It's easy to understand. And I got a good result in hopper.But when I transform it to the other environments like ant or humanoid, it got a bad work.So what parameters should I tune to get a better reward on these environment? Do you have any try? Looking for your reply.Thanks!

toshikwa commented 2 years ago

Sorry for the late response.

In the gail paper, experiments are done with 25M training steps for the most environments including half cheetah, and 75M steps for humanoid. I don't have enough computational resources to test on humanoid env, but at least I have tested on half cheetah with 20M steps and it worked well.

Have you tried on humanoid with 75M steps?