Procedural automation of general game level generation: the good, the bad and the ugly

Procedural automation of general game level generation: the good, the bad and the ugly

Posner, Noah (TU Delft Electrical Engineering, Mathematics and Computer Science)
Krijgsman, Caspar (TU Delft Electrical Engineering, Mathematics and Computer Science)
Baran, Krzysztof (TU Delft Electrical Engineering, Mathematics and Computer Science)
Wisboom, Matthijs (TU Delft Electrical Engineering, Mathematics and Computer Science)
Lambregts, Steven (TU Delft Electrical Engineering, Mathematics and Computer Science)

Bidarra, Rafael (mentor)

Delft University of Technology

2020-04-28

The monumental goal of Artificial Intelligence (AI) is to model general solutions that can be applied to perform a variety of tasks that normally demand human intelligence to solve. Traditionally human game developers painstakingly design and tweak levels until achieving the precise output of their heart’s desire. In the gaming industry, AI for level generation can reduce the need for labour-intensive human design. A general game AI for level generation can only be created once we have a method to describe video games. Video Game Description Language (VGDL) is a high-level language for describing 2D arcade games that consists of two parts, a game and level description. Using this language allows us to analyze games at their mechanical level. The problem of General Video Game Level Generation (GVG-LG) can thus be defined as follows: construct a generator that, given a game (e.g. described in VGDL) which can be played by some AI player, builds any required number of different levels for that game which are enjoyable for humans to play [1]. This research investigates the characteristics of what makes automation of general level generation for 2D video games difficult, identifying what exactly makes it so challenging. Solutions such as algorithmic approaches and design patterns shall be presented. By investigating the techniques that have been used so far, empirical evidence will provide key insight into which techniques are most promising to improve level generation quality in the future.

Algorithms
AI
Design Patterns
Game Level Generation
VGDL

http://resolver.tudelft.nl/uuid:7ce94622-20d9-4239-970b-a758e87e8a2d

Student theses

bachelor thesis

© 2020 Noah Posner, Caspar Krijgsman, Krzysztof Baran, Matthijs Wisboom, Steven Lambregts