Generative Adversarial Networks Based PCG for Games
Nimisha Mittalella, Priyanjali Pratap Singh, Prerna Sharma · 2021
With the steep transition toward automation in virtually every aspect of human lives, content creation is no exception. The automation of content creation, formally called Procedural Content Generation (PCG), has led researchers to investigate novel approaches to create gaming content as well. Minimizing or making human participation optional, the various methods have been investigated over the past few decades. In recent years, however, with the application of artificial intelligence (AI), several cutting-edge technologies have been utilized in pursuit of these goals. Also, the emergence of state-of-the-art machine learning (ML) algorithms, followed by the incorporation of edge-cutting deep learning (DL) methods, such as reinforcement learning, evolutionary computation, and generative adversarial networks (GANs), have sped up the tempo. The superior results achieved by these developments have acted as a catalyst to improvise upon these algorithms and are paving the path to developing revolutionary algorithms in recent times. One such area of interest is the application of GANs for games. This survey explores the contemporary usage of GANs in the gaming domain, addresses the needed developments necessary for further progress, and is accompanied by some possible future developments using these techniques for gameplay content generation. The existing application of GANs in games is also expounded via a concise overview of existing games based on these novel approaches.