Value-based Reinforcement Learning Methods on Breakout Game

Reinforcement Learning · DQN · Q-learning

V3: Reach the highest score of 287

V2: Set up an abstraction to show how to use Q-learning to solve brick game

V1: a tough model which can average eliminate 15 - 20 bricks.

This is an ongoing project, you can see the V1 source codes at: GITHUB_LINK

The V1 version has limited performance on playing this game. I'm thinking to optimize it when I have time. Now I'm busy in learning RL foundamental knowledges, which can help me figure out what happend in a lower level and allow me to modify this model to achieve the final goal: win the game (eliminate all the bricks). The V1 Version's video is showing below:

GitHub