Work

Procedural Dungeon Generation : Comparing Binary Space Partitioning to an extended Cellular Automata with postprocessing: Flood Fill and Room Splitting

Year: 2025
Type: dissertation
Abstract: This thesis compares two Procedural Dungeon Generation (PDG) approaches building on Binary Space Partitioning (BSP) and Cellular Automata (CA) regarding three PDG properties: expressivity, reliability and controllability. To enable a fair comparison, the CA method is extended with two post-processing steps: a flood fill algorithm to detect rooms, and a room splitter to enforce room size constraints. Both methods are then used to generate 2D top-down dungeon layouts within a 50x50 tile map. The evaluation uses two agent-based metrics: Difficulty, how much the agent must explore to find the best path from the start room to the end room. Diversity, how different the agent’s actions are when solving different layouts. The agent uses the A* pathfinding algorithm to navigate the dungeon layouts. The dungeon layouts are represented as graphs, where room are nodes and corridors are edges. A Minimum Spanning Tree (MST) is constructed to ensure room connectivity, and additional corridors are added selectively to increase layout complexity. Each configuration generates 1000 dungeons, and the results are visualized using expressive range analysis via 2D heatmaps of Difficulty and Diversity. The results show that BSP is more reliable and predictable, consistently generating well-structured dungeons. In contrast, CA exhibits greater expressivity, producing a broader and more varied set of layouts, but at the cost of occasional reliability issues, such as isolated rooms or visual artifacts, the causes of which remain unresolved. The choices between BSP and CA depends on design priorities: BSP is preferable for structured and predictable layouts, while CA might be better for scenarios demanding variety and emergent structure.
Author: Strinnholm, Erik
Institution: Umeå University
Language: English
Open Access status: green
OpenAlex ID: W7080052116
Cites: 15
Cited by:
Related to: 10