Study of Artificial Intelligent Algorithms Applied in Procedural Content Generation in Video Games

This study examines how artificial intelligence and algorithmic techniques can be applied to procedural content generation (PCG) in video games, a response to the rising cost of producing game content. Its central aim is to evaluate PCG methods through working prototypes and to assess their practical performance on consumer hardware. The authors built four freely available demonstration games, each illustrating a different technique, and measured generation speeds to test their viability.

The maze-generation experiments, conducted in a Unity-based demo on an Android smartphone, found Prim's algorithm to be the most reliable overall, averaging roughly 1206 ms on 50×50 labyrinths, while Depth First Search performed worst at about 1581 ms. A dungeon generator using space binary partitioning and a seven-step lock-placement method produced guaranteed-solvable layouts, generating dungeons of up to 16 rooms in under one second. Other systems included a grid-based 2D strategy map generator that combines cellular-automata-inspired terrain with A and Dijkstra-based placement, where base placement emerged as the principal bottleneck at roughly 18 seconds on 60×60 maps.

A Minecraft-style 3D terrain demo illustrates chunk-based generation using Perlin noise and seeded height maps to produce coherent, endless landscapes. Performance data here is a key limitation: a 16×16×16 chunk averaged 4.62 seconds, and the 64-chunk initial navigable terrain averaged 10.84 seconds, which the authors acknowledge remains computationally costly.

The authors conclude that no general method yet exists for automatically producing game content, and that procedural content generation is promising but underdeveloped. They anticipate that its growth will be accelerated by industry demand for content and by its combination with artificial intelligence.