King of Games Learning Environment

Requirements:

pip install pybind11
apt-get install clang llvm

Build instructions:

chmod u+x i
./i

To use:

from kogle.kogle import Kogle
env = Kogle(game_name="Breakout", sticky_action_prob=0.1)

The interface is pretty self-explanatory.

• env.step(action) takes a step in the environment and returns the reward.
• env.observation gets the current frame. env.observation_shape gives its shape.
• env.state gets the most recent env.obs_stack frames, concatenated along the first dimension. env.state_shape gives its shape.
• env.legal_action_mask gets a boolean mask of length env.action_count indicating which actions are legal in this state.
• env.legal_action_set gets an integer set containing the legal actions.
• env.terminated gets a boolean corresponding to whether the environment has terminated.

Run python interactive.py to get an interactive version of the game that can be played with arrow keys. (Requires tkinter.) Use spacebar to fire, z for no-op, and WASD to move and fire at the same time.

Run python interactive_live.py to get an interactive version with timed frames. Default 30fps. (Requires pygame.) Uses same controls as python interactive.py.

Run python benchmark_speed.py to measure the FPS.