AI generates test oracles for AR apps with 88% accuracy
New arXiv paper shows LLM agents can auto-create test assertions that catch bugs in AR code
Researchers introduced a context-aware pipeline that uses large language models to generate and refine metamorphic relations (MRs) for testing augmented reality applications — a domain where dynamic interactions between virtual content, physical environments, and code make traditional test oracles difficult to define. Evaluated on a dataset of 142 mobile AR system repositories, the pipeline generated 14,916 candidate MRs across three context configurations, with hierarchical context yielding the broadest coverage (7,004 MRs across 142 repositories and 5,167 class–method pairs) and lower redundancy. An agentic
- Generated 14,916 candidate MRs across 142 AR repositories using repository-aware LLM pipeline
- Hierarchical context reduced redundancy by 79% and achieved 88.2% selection accuracy for context-aware relations
- Refined MRs were directly usable as test assertions and detected real-world code mutations in case study
Why It Matters
Automates test oracle generation for AR apps, cutting manual testing time while improving bug detection reliability