An infinite world generated at runtime in Unreal Engine 5.5 — from multithreaded C++ terrain to biome-aware PCG environments.
The goal was to build a functional prototype that creates varied environments on demand as the player moves — without freezing the game thread or requiring a hand-authored landscape.
The solution separates the problem into two cooperating layers. A custom C++ system produces the underlying terrain and streams chunks around the player. Unreal Engine's PCG framework then interprets biome data and populates those surfaces with vegetation and infrastructure.
Configurable noise layers define large-scale forms and local terrain detail. Each biome can use its own height formula.
Vertex and mesh data are computed on background threads so expensive generation work does not block gameplay.
Terrain is divided into chunks that appear around the player and are removed after leaving the active radius.
Temperature and humidity drive biome weights. Interpolation removes hard seams between neighboring regions.
Modular graphs place vegetation, rocks, roads, fences, buildings and villages using biome-aware rules.
The prototype supports both stylized and photorealistic presentation. The terrain data stays independent from the content rules and art direction layered on top.
Generation was evaluated while the player moved at high speed and chunks continuously streamed in and out. The thesis compares Procedural Mesh Component and runtime Static Mesh approaches across several resolutions and world sizes.
Procedural Mesh average build time, compared with 2.9 s for runtime Static Mesh.
Procedural Mesh average build time, compared with 8.6 s for runtime Static Mesh.
Procedural Mesh stabilized around this level in the documented test; Static Mesh held approximately 90 FPS.
Measured on the thesis test configuration. Results are evidence for the prototype, not a universal engine benchmark.
I also created a small educational application for experimenting with procedural noise. It makes it easier to see how individual noise functions and their combinations shape the height data later used by the terrain generator.