Automated Testing of Agent-Driven Construction in 3D Environments

Abstract

Automating agents to build structures in video games is an underdeveloped topic that could provide many applications in automated video game playtesting as well as improving artificial intelligence for non-player characters. The most important component of such a system is a planning algorithm that sequences the order in which pieces are placed to prevent mistakes or unsolvable scenarios. We create such a construction planning system by expanding on the existing iv4XR framework. We use the video game Space Engineers to show that our method works and that there exist many valid ways of sequencing with varying levels of efficiency. We conclude that a depth-first approach provides the most performant base but that the addition of situational awareness factors greatly help with consistency and robustness.