<![CDATA[User interface (UI) test automation on Android frequently breaks when developers rename element attributes during refactoring, rendering previously valid locators unresolvable and imposing significant maintenance overhead. Existing self-healing approaches predominantly target web DOM and lack post-action validation, risking false healing where a wrong element is silently accepted. This study introduces AURA, a runtime self-healing layer for Appium-WebdriverIO that chains five deterministic recovery strategies, a widget-family post-action validator, and an optional machine-learning reranker. A controlled benchmark comprising 490 refactoring scenarios across five synthetic Android applications and six mutator types demonstrates that AURA achieves a 99.39% correct action rate with only 0.61% false-healing rate, significantly outperforming the adapted Similo baseline (95.71% / 4.29%) at p < 0.0001 (McNemar exact test). External validation on six production Google Android applications (130 scenarios) confirms a 100% correct rate with a bounds-IoU enhanced validator. Cache learning reduces per-find latency by 95.1% from the second session onward.]]>
Copyrights © 2026
