Language

▪ Python

Status

• Completed 2019-11-28

With the help of the reinforcement learning method Deep Q Networks, the game Tetris was implemented. This project was completed together with Emilie Ho and Ronja Faltin.

The goal was to train an AI to play and beat the game as a human would. The agent was trained by interacting with the environment. The first thing that happened in the process of training the agent, was that the agent was given a state from the environment. Depending on the state, the agent chooses an action to send back to the environment. Actions in return give rewards which can be positive, negative or zero. For this specific project, the actions were to move a brick left, right, rotate a brick or instant drop a brick. The goal was to train the agent to maximize the total reward it collects over an episode. An episode was one game from beginning to game over. For each episode, the agent learned which actions provided positive rewards and which actions to avoid.

You can checkout the project on Github!