Grammar-based Procedural Content Generation from Designer-provided Difficulty Curves


In experience-driven procedural content generation (EDPCG) [15], the challenge of (parts of) a level are often subject to player-adaptive optimisation. However, this may interfere with the design goals of the game designer with respect to the difficulty build-up of the entire level, e.g., the designer may have specific ideas about where the climax of a level ought to be. This can be a reason for designers to not adopt experience-driven procedural techniques. In this paper we mitigate this, by meeting the designers half-way: the designers provide a set of allowed difficulty curves for a level, and decide where the AI is allowed to switch between these, e.g., after each or certain segments or only between levels. This way, the designer is in control of the tension levels and ‘feel’ of the level, while still allowing player adaptivity. This paper describes how to generate level(s) (segments) using difficulty curves, and how this can be applied to experience-driven procedural content generation. Experiments that validate our approach in an actual, open-source action-adventure game, reveal that it is consistently able to generate entire game levels that closely approximate distinct difficulty curves. Also, the adopted generative grammar approach ensures that the generated content will never be unplayable, as it results strictly from (presumably adequate) designer-provided grammars. Finally, the obtained experimental results show that the procedural generation of game levels consistently takes place in a reasonably computationally efficient manner. Given these obtained results, we conclude that our enhanced procedural approach provides an effective basis for generating game levels according to designer specifications, yielding new options for PCG.

Extracted Key Phrases

4 Figures and Tables

Cite this paper

@inproceedings{Traichioiu2015GrammarbasedPC, title={Grammar-based Procedural Content Generation from Designer-provided Difficulty Curves}, author={Mircea Traichioiu and Sander Bakkes and Diederik M. Roijers}, booktitle={FDG}, year={2015} }