Gym-style implementation of the original dm_env style Ballet environment which is introduced in Towards mental time travel: a hierarchical memory for reinforcement learning agents.
- Gym-style implementation of the original dm_env style Ballet environment
- Support gymnasium>=0.27.0
- Implement wrappers for gym-balletenv:
- And the following gymnasium wrappers are tested for gym-balletenv:
git clone https://github.com/jinPrelude/gym-balletenv.git
cd gym-balletenv
pip install -e .Example code for gym-balletenv :
from gym_balletenv.envs import BalletEnvironment
from gym_balletenv.wrappers import GrayScaleObservation
env = BalletEnvironment(env_id="2_delay16", max_steps=320)
env = GrayScaleObservation(env)
obs, info = env.reset(seed=0)
terminated = False
while not terminated:
action = env.action_space.sample()
obs, reward, terminated, truncated, info = env.step(action)Check out the repo gym-balletenv-example more usage of gym-balletenv and its wrappers.
Easy mode is a custom feature that removes various color options and limited dancer shape by the number of dancers. So all the colors of the dancers are "red", and only 4 shape of the dancers will be sampling for 4 dancers environment. This feature is made for rapid model evaluation purpose.
Simply add "_easy" at the end of the env_id :
from gym_balletenv.envs import BalletEnvironment
env = BalletEnvironment(env_id="2_delay2_easy", max_step=320)