Risk-Adaptive Cloud-to-Edge Application and Data Update Architecture for Android-Based Embedded Systems Using Lightweight Edge AI Validation

Abstract

Android-based embedded systems increasingly receive application packages, configuration data, policy files, machine-learning models, and security metadata from cloud services. These cloud-to-edge updates improve product maintainability, but they also expand the attack surface: a device can receive a stale manifest, a mismatched split package, a corrupted artifact, an unsafe policy revision, or a model update that is technically valid but operationally risky for the current device state. This paper proposes a risk-adaptive cloud-to-edge application and data update architecture for Android-based embedded systems using lightweight edge AI validation. The architecture combines signed update manifests, artifact-level integrity checks, application data-security controls, persisted update-state management, package compatibility verification, rollback-safe orchestration, and a quantized on-device risk classifier. The edge AI component does not replace deterministic security gates; instead, it prioritizes, defers, quarantines, or escalates updates based on contextual risk features such as signature status, manifest freshness, package version drift, retry history, rollback frequency, network trust, power state, storage pressure, and prior installation failures. The proposed architecture is designed for industry environments where Android applications run on embedded, automotive-style, kiosk, retail, industrial, or managed-device platforms and where cloud-delivered updates must be secure, observable, recoverable, and operationally safe. A reproducible evaluation framework is provided using metrics for false acceptance, false rejection, validation latency, memory overhead, rollback detection time, manifest replay resistance, and update completion reliability.