A Prototype Debug AI by Rule-based Game Screen Recognition for Tetris

Abstract

To build a Debug AI that is capable to discriminate whether or not something is a bug instead of human debuggers, there are many technical problems such as bug discovery from play movies and automatic generation of a sequence of play operations for efficient bug discovery. To solve these problems, rule-based and/or machine learning methods would be able to be applied. Therefore, as the first step, this paper develops a prototype debug AI for Tetris as a target game to discovery bugs by recognition of game screen images in a play movie by a human player, which may have bug(s), and rule-based discrimination on whether or not each image is a bug, and validates its performance.