Build an AI to play Tetris. The team which clears the most rows before losing wins.
First Place. The AI was shut off after clearing 50,000 moves in the competition. It made it to over 1 million moves in testing.
Every possible rotation and position of current+next piece were placed on the board and the “badness” of the board was rated. The board with the lowest badness was selected and the current piece placed in that position. Badness ranking was determined by measuring a variety of factors (ex mean column height, number of holes, column height std dev) multiplied by constants. These constants were tuned using a genetic algorithm running on a laptop for a few days.
Source Code: Github