The reinforcement learning Pong game demo written in javascript, which runs in web browser.

Two AI agents play the game. You only can be act as an audience.

If you are a programmer:

1. install VSCode and Live Server extension
2. open the index.html file with Live Server extension

If you are not a programmer: deploy the project as an app in any one HTTP server

Pong has the action space of 2 with the table below listing the meaning of each action’s meanings

Pong’s state is a tuple with 5 items. the table below lists the meaning of each item meanings

the x positive direction is to the right the y positive direction is to the up

You get the reward score when the ball pass the paddle or collide with the paddle.

reward = math.log(abs(paddle_pos - ball_position.y) / area_height + 0.000001)

• paddle_pos is the paddle center y position
• ball_position.y is the ball center y position
• area_height is the game area height

Please refer to the training README.mdfor training details. How to train

1. the training screen shot

2. the game screen shot