Internal Documentation

RL Project Documentation

init.py

Where the playing modes are defined. The playing modes are (as explained on Github): WimblepongVisualMultiplayer-v0: Two agents are playing against each other based on the pixel observation • WimblepongVisualSimpleAI-v0: One agent plays against a SimpleAI based on the pixel observations • WimblepongMultiplayer-v0: Two agents are playing against each other based on the absolute positions of the ball and the paddles • WimblepongSimpleAI-v0: One agent plays against a SimpleAI based on the absolute positions of the ball and the paddles

simple_ai.py:

Where the Ai is defined. Question: What do we need to in reset() ?

wimblepong.py

Class Rect
	Defines rectangle → Question:This represents the ball, right?

Class Ball
	Defines the ball in the game

Class Player
	Defines the player with his/her paddle.

Class Wimblepong
	Where the UI is rendered.

test_pong_ai.py

Actions are always zero because there is no agent

test_pong_ai_multi.py

Same as test_pong_ai but there are two agents here

Some ideas

As suggested in the blogpost we could implement a neural network to computer the probabilities of moving UP, DOWN or STAY. If we manage to implement parallel data collection, we could collect data about the ball and our opponents paddle