Enhancing automatic level generation for platform videogames

• Autor(a) principal: Mourato, Fausto José da Silva Valentim
• Data de Publicação: 2015
• Idioma: eng
• Título da fonte: Repositório Científico de Acesso Aberto de Portugal (Repositórios Cientìficos)
• Texto Completo: http://hdl.handle.net/10362/18497
• topic: Procedural content generation; Game design patterns; Platform videogames; Human factors; Domínio/Área Científica::Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática

This dissertation addresses the challenge of improving automatic level generation processes for plat-form videogames. As Procedural Content Generation (PCG) techniques evolved from the creation of simple elements to the construction of complete levels and scenarios, the principles behind the generation algorithms became more ambitious and complex, representing features that beforehand were only possible with human design. PCG goes beyond the search for valid geometries that can be used as levels, where multiple challenges are represented in an adequate way. It is also a search for user-centred design content and the creativity sparks of humanly created content. In order to improve the creativity capabilities of such generation algorithms, we conducted part of our research directed to the creation of new techniques using more ambitious design patterns. For this purpose, we have implemented two overall structure generation algorithms and created an addi-tional adaptation algorithm. The later can transform simple branched paths into more compelling game challenges by adding items and other elements in specific places, such as gates and levers for their activation. Such approach is suitable to avoid excessive level linearity and to represent certain design patterns with additional content richness. Moreover, content adaptation was transposed from general design domain to user-centred principles. In this particular case, we analysed success and failure patterns in action videogames and proposed a set of metrics to estimate difficulty, taking into account that each user has a different perception of that concept. This type of information serves the generation algorithms to make them more directed to the creation of personalised experiences. Furthermore, the conducted research also aimed to the integration of different techniques into a common ground. For this purpose, we have developed a general framework to represent content of platform videogames, compatible with several titles within the genre. Our algorithms run over this framework, whereby they are generic and game independent. We defined a modular architecture for the generation process, using this framework to normalise the content that is shared by multiple modules. A level editor tool was also created, which allows human level design and the testing of automatic generation algorithms. An adapted version of the editor was implemented for the semi-automatic creation of levels, in which the designer may simply define the type of content that he/she desires, in the form of quests and missions, and the system creates a corresponding level structure. This materialises our idea of bridging human high-level design patterns with lower level automated generation algorithms. Finally, we integrated the different contributions into a game prototype. This implementation allowed testing the different proposed approaches altogether, reinforcing the validity of the proposed archi-tecture and framework. It also allowed performing a more complete gameplay data retrieval in order to strengthen and validate the proposed metrics regarding difficulty perceptions.