Automated level generation and difficulty rating for Trainyard

Abstract

This thesis presents a framework for automatic level generation for the puzzle game Trainyard. We made a replica of this game, called Flight, and integrated the level generation framework into it. This framework is able to automatically generate levels of various difficulty. The generation of levels is divided into three components: (i) the level generator, which creates the level, (ii) the level solver, which checks the feasibility of the generated level, and (iii) the difficulty estimator, which rates the difficulty of the generated level.

To test the presented framework we did two user studies: (i) a pilot study, which tested whether the difficulty ratings of the original game1 correspond to the ratings given by the player, and (ii) a full user study, which tested whether the assigned difficulty ratings of the generated levels corresponds to the player’s experience.

The results of the level generation framework are very promising, levels are automatically generated, tested and rated successfully. In further research the process could be optimized to make the level generation faster and generated levels more user specific.