Improving Human Players' T-Spin Skills in Tetris with Procedural Problem Generation

This research explores the application of procedural content generation to create educational puzzles for Tetris, specifically targeting the mastery of the T-spin technique. The primary thesis posits that automatically generated, two-step T-spin problems can effectively train human players, improving their ability to execute this complex maneuver during competitive gameplay. The study utilizes a reverse-generation method to create these puzzles, incorporating specific features such as tetromino connectivity and spatial positioning to predict problem difficulty and interestingness.

The methodology involved a two-stage experiment conducted with fifteen male participants, all of whom were interested in gaming. In the first stage, participants solved 50 two-step T-spin problems, providing ratings that were used to train a LightGBM-based predictor for difficulty and interestingness. The resulting models demonstrated high predictive accuracy, with a symmetric mean absolute percentage error of 8.74% for interestingness. In the second stage, participants were divided into three groups—a control group, a group trained with random problems, and a group trained with interesting problems—to measure skill improvement through competitive play against CPU agents.

Key findings indicate that while all groups showed increased win rates, the groups trained with T-spin problems exhibited a significant rise in T-spin frequency, with growth rates of 212% and 334% compared to a decline in the control group. Although the increased focus on T-spin occasionally led to slower gameplay speeds and minor execution errors, the results confirm that procedural problem generation is a viable tool for skill acquisition. The study concludes that personalized training systems can successfully enhance specific player techniques, suggesting that future research should explore multi-step problems and additional advanced Tetris maneuvers.