Procedural Content Generation (PCG) allows game developers to create worlds in games easier compared with creating them manually. One of many PCG algorithms is Cellular Automata (CA). CA was used in this research together with Poisson Disk Sampling (PDS) to generate two-dimensional dungeons in a roguelike game. PDS was modified to adapt to a grid-based environment to increase computation's efficiency. Each room generated in the game was connected to each other. Together, all the rooms created a dungeon. The connections between the rooms were architected using Depth First Search (DFS) algorithm as commonly used in maze generation. The created dungeons became the world in the game. Players could explore the dungeons to conquer them and collect resources. Then, players could spend the resources to upgrade their character. According to the research evaluation, using only CA was not sufficient to generate dungeons with specific designs. Certain pipelines were demanded to adjust and to control the end results. The algorithm still run adequately fast, which was around 73 milliseconds, despite all the pipeline additions.