ARIA Splits Infotainment Test Automation Across Four Agents Per Step Instead of Overloading One

arXiv 2609.04913·low signalXBlueskyLinkedInCopy link

arXiv 2609.04913 targets Android automotive infotainment validation, which still relies on manual testing that does not fit agile releases and OTA updates, while scripted automation couples test logic to implementation and produces brittle suites. ARIA runs end-to-end tests through visual interaction using a closed-loop pipeline of four specialized agents per step plus a report stage, taking single-sentence scenarios of path, action and expected outcome as input. The design argument generalizes: the authors attribute existing single- and dual-agent web/mobile frameworks' hallucinations and unproductive exploration loops to loading perception, planning, action selection and validation onto one or two models.