A lightweight wrapper around the DeepMind Control Suite that provides the standard OpenAI Gym interface. The wrapper allows to specify the following:
- Reliable random seed initialization that will ensure deterministic behaviour.
- Setting
from_pixels=True
converts proprioceptive observations into image-based. In additional, you can choose the image dimensions, by settingheight
andwidth
. - Action space normalization bound each action's coordinate into the
[-1, 1]
range. - Setting
frame_skip
argument lets to perform action repeat.
pip install git+git://github.com/denisyarats/dmc2gym.git
import dmc2gym
env = dmc2gym.make(domain_name='point_mass', task_name='easy', seed=1)
done = False
obs = env.reset()
while not done:
action = env.action_space.sample()
obs, reward, done, info = env.step(action)