VRExplorer boosts VR testing coverage by 122.8% with AI agent
New model-based tool outperforms state-of-the-art VRGuide significantly.
With VR applications growing rapidly in complexity, ensuring their quality becomes a critical challenge. Traditional testing tools fall short due to the unique 3D environments, diverse object interactions, and intricate task sequences in VR. To address this, researchers from Sun Yat-sen University and other institutions present VRExplorer, a semi-automated testing approach that models VR interactions generically via the Entity, Action, and Task (EAT) framework. This framework allows the system to understand and interact with any virtual object, making it highly adaptable to different VR scenes.
Built on the EAT framework, VRExplorer deploys an intelligent agent that combines Unity’s NavMesh for efficient navigation with a Probabilistic Finite State Machine (PFSM) to systematically execute interactions. In extensive experiments across 11 representative VR projects, VRExplorer consistently outperformed the previous state-of-the-art tool VRGuide, achieving up to 122.8% improvement in executable lines of code (ELOC) coverage and 52.8% improvement in method coverage. More importantly, the tool successfully identified two functional bugs and one non-functional bug in real-world VR applications. The paper, accepted at ASE 2025, demonstrates a significant step forward in automated VR quality assurance.
- VRExplorer uses a novel EAT framework (Entity, Action, Task) to generically model any VR interaction.
- Achieves up to 122.8% higher ELOC coverage and 52.8% higher method coverage than existing tool VRGuide.
- Successfully detected 2 functional and 1 non-functional bugs in real-world VR projects.
Why It Matters
Automates complex VR testing, boosting coverage and catching bugs, which accelerates development and improves user experience.