Scale-Free Evolutionary Level Generation
André Siqueira Ruela, Karina Valdivia Delgado · 2018
This work proposes a new approach for the graph-based procedural level generation for games, through evolutionary algorithms. The levels are encoded as a graph structure inspired by the concepts of a scale-free network. A scale-free network lacks an internal scale as a consequence of the coexistence of nodes with different degrees in the same network. This approach aims to avoid the generation of linear, repetitive and grid-like levels, giving the algorithm additional freedom to explore the search space for diverse solutions. The algorithm was designed to provide a smooth mixed-initiative authorial control, allowing the designer to adjust constraints parameters, input aesthetically desired properties, manage critical contents of the level and even edit edges and nodes in a drag-and-drop manner. The results show that the algorithm can evolve scale-free structures with moderated nonlinearity, which is taken as an ideal measure for game levels that feature player progression like lock-and-key puzzles.