The above AI is a demonstration of the DQN (Deep Q-Network) I built. I wrote it from scratch in Objective C so I can learn the detail of how DQN works, but also to have it be able to play the game I created for fun.
The AI is trained over 4 days, playing the game over and over. And, it’s able to achieve superhuman level and reached a high score of 250 (my personal best is around 80). The AI is showing no sign of slowing down even after 4 days.
If you think you can beat the AI, you can try the game Jumpy Ball. The neural network source code can be found here.
