Presentation Title

Causal Inference in Procedural Dungeon Generation

Faculty Mentor

Andrew Forney

Start Date

17-11-2018 8:30 AM

End Date

17-11-2018 10:30 AM

Location

HARBESON 14

Session

POSTER 1

Type of Presentation

Poster

Subject Area

engineering_computer_science

Abstract

Procedural content generation (PCG) is a common technique used in modern games to randomly vary the creation of production-heavy assets used during development. One common application of PCG is to autonomously create diverse game environments from simple specifications. Previous PCG techniques have specified associational relationships between environmental features to vary their outputs, which are then difficult to modify post-production. Recent developments in causal inference have provided a richer toolset beyond simple associations that give greater control over a probabilistic model’s expressiveness. In this project we apply these tools to procedural dungeon generation, in which users may specify not only associational, but also causal and counterfactual generation criteria. The result is a framework called DunGen that allows for more finite control over the output both before and after generation, the result of which is a blueprint-style image displaying room layouts and adjacencies.

This document is currently not available here.

Share

COinS
 
Nov 17th, 8:30 AM Nov 17th, 10:30 AM

Causal Inference in Procedural Dungeon Generation

HARBESON 14

Procedural content generation (PCG) is a common technique used in modern games to randomly vary the creation of production-heavy assets used during development. One common application of PCG is to autonomously create diverse game environments from simple specifications. Previous PCG techniques have specified associational relationships between environmental features to vary their outputs, which are then difficult to modify post-production. Recent developments in causal inference have provided a richer toolset beyond simple associations that give greater control over a probabilistic model’s expressiveness. In this project we apply these tools to procedural dungeon generation, in which users may specify not only associational, but also causal and counterfactual generation criteria. The result is a framework called DunGen that allows for more finite control over the output both before and after generation, the result of which is a blueprint-style image displaying room layouts and adjacencies.