Procedural Generation of Dungeon Maps, Missions and Rooms

• Ano de defesa: 2018
• Autor(a) principal: Pereira, Leonardo Tortoro
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Adaptive game content; Algoritmo evolutivo; Conteúdo adaptativo de jogo; Evolutionary algorithm; Geração de missão; Geração de nível; Geração procedural de conteúdo; Level generation; Mission generation; Procedural content generation
• Link de acesso: http://www.teses.usp.br/teses/disponiveis/55/55134/tde-25032019-144917/

The present research proposes two procedural content generation approaches for missions and play space in a game with dungeons, and a game prototype developed as a proof of concept for the feasibility of the algorithm and as a test framework for user-based experiments. The first approach will define missions by generating dungeon maps together with locked doors and keys through the use of an evolutionary algorithm. The second approach will populate the generated dungeon space by filling the content of dungeon rooms using space-filling curves and cellular automata algorithms. The evolutionary algorithm evolves tree structures encoding information about the dungeon. The goal is to converge the generated dungeons as close as possible to satisfy the set of parameter inputs provided by a game designer. The dungeon holds information about the amount of rooms, the connections between them and their position in a 2D map. There are also relevant semantic information in the content for the generation of narrative, which presents itself by the placement of unique pairs of keys and locks throughout it. Thus, a feasible way to finish the dungeon can be procedurally generated. The content of rooms are generated using space-filling curves algorithms such as Hilbert and Moore curves as well as Conways Game of Life Cellular Automata. Computational results report that the evolutionary algorithm provides dungeons with up to a 100 rooms very close to the desired ones for a range of different parameter inputs. The early validation tests with humans show no statistical difference between levels procedurally generated and those made by humans. Further user-centred validation tests with the game prototype show the algorithm-generated levels are perceived as equally or even more human-made than their human-authored counterparts, as well as funnier and more difficult. Thus, the research aims to generate gameplay elements combining different algorithms for a single solution, which could be easily adapted to a range of different games.