Atari simulator using python lib gym
Due to the dependency opencv, the program is developed under PYTHON2.
This package depends on the following packages:
First, create your cfg file via:
python2 gen_cfg.py cfg/<some_file_name>.py <game_name>
where <game_name> is the game’s name in OpenAI’s Gym. (e.g. Enduro-v0)
In this step, you need to follow the instructions and create the key mapping for your game.
E.g. when the program asks you to input a key for ‘DOWN’ action, you may probabily press ‘
A tricky thing is that after mapping all game-related actions, you need to map a special action ‘quit’, which can be used to stop the game while playing.
Now it’s time to play the game. Use:
python2 main.py cfg/<some_file_name>.py --fps <fps>
Note at any time you can use ‘quit’ action to quit your play. After you press the ‘quit’ button or when the game normally ends, a replay pickle will placed in the replay folder.
One tricky thing is that you can adjust the fps for the game if you want to play it in a slower mode.
The replay file can be loaded by module pickle without any dependency (oh, numpy is needed). The only tricky thing in the replay file is that for easier usage, the observation in each ‘step’ actually refer to the last observation. That is, for a dict in replay['steps'], action (and reward) is the action (and reward) human expert plays after seeing observation.