Title

BlockBot: A Neural Network That Plays Tetris

Team members

Elijah Swearingen (eswearin), Sophia Cooney (cooney7), Tyler Dierckman (tdierckm)

Project description

The goal of our project is to train AI to play Tetris. We plan to build our own dataset by writing a script that can analyze online videos of skilled tetris players and extract their moves along with the current state of the board. To keep things simple, we will initially ignore the pieces "up next" and focus strictly on information contained on the board itself. The dataset will consist of a bunch of board setups (both fixed pieces and falling piece) along with the next move made by the human player as determined by our script. We will train the model using a snapshot of a board setup as input to our model, and the "gold standard" move that the human made in this scenario as the target. For example, one dataset entry might be an empty board with a falling piece at the top, and a corresponding target of "down," meaning that the human player took no action and let the piece fall by one block. We will consider the options to move left, move right, move down (let the piece fall), and rotate. Our dataset will represent the "correct" actions to take given a whole bunch of scenarios. The plan is to train the model to make one move of the falling piece at a time based on the current state of a board. Hopefully, after the model is trained on what actions it should take given a board state, we can recursively run forward inference on a developing game in order to "play" tetris. We will likely have to code a simple version of tetris ourselves or find an open source version that we can interface with to demonstrate our model "playing." If we get the model to successfully play Tetris while focusing on 1 block at a time, we have a more extreme secondary goal of having the AI use information on the blocks coming up to factor in its placement of each block. But to start, we will focus on placing 1 block at a time with only the current board's information/setup.

tdierckm/BME450-project

About

Resources

Stars

Watchers

Forks

Releases

Contributors 3