sgoodfriend's picture
PPO playing impala-BreakoutNoFrameskip-v4 from https://github.com/sgoodfriend/rl-algo-impls/tree/e47a44c4d891f48885af0b1605b30d19fc67b5af
b05c680
raw
history blame contribute delete
618 Bytes
import gym
import gym.spaces.dict
import numpy as np
from gym import ObservationWrapper
class GetRgbObservation(ObservationWrapper):
def __init__(self, env) -> None:
super().__init__(env)
assert isinstance(env.observation_space, gym.spaces.dict.Dict)
self.observation_space = env.observation_space["rgb"] # type: ignore
if getattr(env, "is_vector_env"):
self.single_observation_space = env.single_observation_space["rgb"] # type: ignore
def observation(self, observation: gym.spaces.dict.Dict) -> np.ndarray:
return observation["rgb"] # type: ignore