search

Denys88 / rl_games

RL implementations
MIT License
800 stars 135 forks source link

Getting error while testing #262

Closed qureshinomaan closed 7 months ago

qureshinomaan commented 8 months ago

I am able to train policies. But when loading the checkpoint to 'play' I am getting the following error : AttributeError: 'Envpool' object has no attribute 'render'

The command I am running is : python3 runner.py --play -c runs/Walker2d-v4_envpool_30-07-00-39/nn/last_Walker2d-v4_envpool_ep_1000_rew_5242.4565.pth --file ./rl_games/configs/mujoco/walker2d_envpool.yaml

I tried searching for a solution and it looks like a problem with gym.

I am not sure if adding time is the best or recommended way?

Denys88 commented 8 months ago

Thats a reason why I switch to the gym version of the mujoco to use render. Could you set render to false in the config? Currently it is true/ player: render: True AFAIK envpool doesn't support rendering at all.

qureshinomaan commented 8 months ago

Thanks a lot for the reply! Any workaround?

Denys88 commented 8 months ago

If you want to render your policy you can reuse same checkpoint but with config with regular mojuco gym. And don't forget set render:True in yaml.

fnuabhimanyu8713 commented 4 months ago

Hey @Denys88 do you have support for the opensource mujoco. Currently, the player calls gym.envs.mujoco.mujoco_env, which calls mujoco_py (old mujoco). How can I call the latest version of mujoco, which is not dependent on gym