SMA‑AutoUI introduces a novel approach to Android UI automation by combining ADB, computer vision, and large language models.
This triple‑layer architecture reduces reliance on fragile element locators and improves test resilience.
The ADB bridge layer provides low‑level device control with minimal latency.
Vision‑based detection uses template matching and lightweight ML models to locate UI elements by appearance.
The LLM cognitive layer interprets natural‑language test steps and triggers self‑healing when UI changes.
Self‑healing logs confidence scores and alternative matches for transparency and debugging.
Early adopters report up to a 70% reduction in locator maintenance after UI refreshes.
The framework works on production‑signed APKs, requiring only USB debugging or wireless ADB.
Installation is straightforward via PyPI with a minimum Python 3.12 requirement.
A fluent API lets writers express tests in near‑plain‑English steps.
MIT licensing encourages community contributions and custom vision model extensions.
Overall, SMA‑AutoUI aims to lower test flakiness and accelerate release cycles for dynamic mobile apps.