Based on the standard OpenAI AtariEnv but returns a tuple observation (image, ram). The image observation is typically a np.array((210, 160, 3), dtype=np.uint8) ram is an np.array((128,), dtype=np.uint8).
Clone this repo and install with pip install -e . from within the cloned directory.
The environments can be used like:
In [1]: import gym
In [2]: import gym_air
In [3]: env = gym.make('MontezumaRevengeDeterministic-image-ram-v0')
[2017-01-24 07:25:18,428] Making new env: MontezumaRevengeDeterministic-image-ram-v0
In [4]: env.observation_space
Out[4]: Tuple(Box(210, 160, 3), Box(128,))
In [5]: env.action_space
Out[5]: Discrete(18)
In [6]: