Procedural Level Balancing in Runner Games

Balancing a game is a long process and relies mainly on subjective feedback from human testers and selective interpretation from game developers. As a first step for completely automate game balancing, we propose a methodology to algorithmically choose features and calibrate their parameters for the procedural level generation of a simple runner game based on testers' feedback.This methodology is used in a 30 seconds game demo with survey and each playthrough is recorded and fed to a reinforcement learning algorithm. We show that the average fun grade steadily grows, proving the effectiveness of the proposed method. The collected data can be further analysed for insights on new features and other major changes.