Augmenting Exploratory Testing Agents for 3D Software via Imitation Learning

Abstract

Recently, due to the emergence of the concept of metaverse, software development that provides various interactions based on 3D virtual space is increasing explosively. Accordingly, the need for 3D software testing research is increasing as well. In order to sufficiently test the interaction between various objects in 3D space, exploratory testing using experts can be effectively applied. However, it is difficult to conduct sufficient testing in the field as the cost of human resources and effort is very high. In order to solve this problem, studies have been conducted to apply the exploratory characteristics of reinforcement learning agents to 3D software testing such as games. In this paper, we propose an exploratory testing automation method in which reinforcement learning agent effectively imitate the expert’s testing behaviors using imitation learning. The experimental results show that the agent applied with imitation learning shows better performance than the existing curiosity-based reinforcement learning agent in terms of defect detection and cumulative reward, and can be effectively used for exploratory testing.