Automated Game Testing based on Bayesian Optimization and ML Agents

Video attached to the paper:
"Bayesian Optimization-based search for Agent control in automated game testing"
Carlos Celemin
Conference on Games, IEEE, 2024.

In this study, we introduce an automated testing approach that employs agents controlling game characters to detect potential bugs within a game scene. To maximize the likelihood of discovering issues, these characters are tasked with efficiently exploring the full map.
Drawing inspiration from successful results of Bayesian Optimization (BO) in various domains, our strategy harnesses the power of BO to execute sample-efficient search.
BO determines the next sampling point by analyzing the collected data so far and calculates the data point that will maximize information acquisition.
While BO is typically implemented using non-parametric models due to their smoothness and ability to model uncertainty, these models encounter scalability issues when applied to large datasets, such as those generated when agents explore game scenes.
To address this, this work introduces a game testing-specific model built on top of a grid map.
This model maintains the desirable smoothness of the traditionally used non-parametric models, enabling extrapolation from gathered observations, and incorporates an uncertainty measure.
%Importantly, its complexity scales linearly with respect to the size of the map and not with the amount of collected data, enhancing scalability.
Our experiments demonstrate that our approach significantly improves map coverage capabilities in both time efficiency and exploration distribution. Furthermore, it has proven twice as efficient at detecting simulated bugs compared to strategies without such sequential optimization.