search

katerakelly / oyster

Implementation of Efficient Off-policy Meta-learning via Probabilistic Context Variables (PEARL)
MIT License
472 stars 125 forks source link

Is mujoco 1.3.1 really necessary? #20

Closed pengzhenghao closed 3 years ago

pengzhenghao commented 4 years ago

Hi there, I am not sure whether the repo rand_param_envs explicitly change anything about gym or mujoco. Now I can run the code, by simply changing the script from:

from rand_param_envs.gym import utils

to

from gym import utils (my gym is the latest one)

and so on, and the code is running smoothly.

So may I ask do you change anything about gym, mujoco_py, MujocoEnv from the original code? If not, then I think it's ok for me to use the official version of gym and mujoco (so that I don't need to install mujoco131..)

Thanks!

katerakelly commented 3 years ago

I believe that rand_param_envs relies on some aspect of MuJoCo 131 and earlier versions to properly implement the joint parameter randomization for the Walker and Hopper experiments. However, I don't know what exactly that aspect is. At any rate, I haven't checked if the parameters actually change the way we would expect if using a later MuJoCo version, so it would be worth verifying that if you use a later version.