Autonomous Anomaly Detection in Games

Abstract

In game development, a lot of time is spent on making sure the product runs smoothly and without too many bugs. This makes testing a very important part of the game development cycle. Delivering a high quality product is critical to maintaining customer satisfaction. Generally, testing games is done by a manual process which can be very time consuming and expensive for developers. In this paper we explore techniques to enable autonomous bug detection and improve the testing pipeline. To achieve this, we employ state-of-the-art anomaly detection techniques. We developed a generic framework for using anomaly detection to analyze the relations between different variables in games. Further, we perform a case study comparing different state-of-the-art anomaly detection algorithms in a wide range of scenarios.

With our framework it is possible to autonomously detect 15\%-20\% of the inserted anomalies with an accuracy of about 90\% without the need for any training data. This research lays the groundwork for easy integration of autonomous testing in games. However, improvements can still be made with future research.