PongAI: Reinforcement Learning to Create Optimal Strategies

To see a full explanation visit my Github repository.

Alternatively, visit the website to see the optimal cups to hit at every step of the game.

The reasoning behind doing this project is two fold. One, Pong is fun and relatable. Two, it’s actually a good candidate for using Q-Learning to build out a policy. The state space is fairly limited, so building out a table doesn’t get to be ridiculous and games can be played really quickly. Also, rewards are given for hitting a cup rather than ending the game, so training can go much quicker.