Block-based Approach to Observe Game Content Space in Endless-runner Game

Game genres are rapidly growing with the technological advances, one of them is the endless runner that becomes favourable in recent years. One example is Flappy Bird which is a fun and popular game. Unfortunately, randomly generating obstacles for Flappy Bird game often frustrates the players due to there were no specific rules to differentiate the difficulty of an obstacle to another. For example, by chance, player A can complete 10 easy obstacles, while player B gets 10 obstacles with a high degree of difficulty. This causes unfair experiences/achievements shared on social media. Meanwhile, a good game should provide distinct difficulty playable from its elements. These problems pose a challenge to make the Flappy bird game have acceptable difficulty levels relevant to its obstacles. In this research, procedural content generation (PCG) technique is proposed to categorize the difficulty of obstacles in Flappy Bird. It combines search-based and learning-based PCG techniques. Search-based PCG is used to define the game's representation relative to a block of obstacles. Then, a bot observes and evaluates them to label the obstacle blocks. Consequently, learning-based PCG acts as the classifier to the observed obstacle blocks with more than 70% accuracy of the selected classification algorithm.