Task-Aware Digital Twin State Updates in Resource-Constrained Edge IoT

Digital twins at the edge require timely state updates under limited communication and computation resources. We propose a task-aware method that selects the state component, representation detail, and sensor/edge processing location under explicit budgets. The selector combines age-based anti-starvation reservations, task-aware ranking, and a proactive check of future update feasibility. For the controlled comparison, a common 180-cycle ESP32 reference trace was used (179 evaluation cycles). The algorithm was evaluated through experimental data replay under six operating conditions, covering two application task profiles and three levels of resource constraints. Under the critical event-monitoring budget, adaptive selection improved shock-event F1 from 0.905172 (round-robin) to 0.967742, with higher temperature and humidity errors. Under critical comfort monitoring, it reduced temperature and humidity MAE to 0.005922 and 0.016313 versus 0.007877 and 0.036536, respectively, but reduced shock-event F1 to 0.742049. A one-cycle exhaustive oracle matched the greedy second-stage objective in 1074 evaluated decisions. Three additional three-hour physical runs (15,229 reference cycles) produced temperature MAE of 0.014282–0.017701 °C and shock-event F1 of 0.968166–0.975199. Freshness violations observed during HTTP communication failures show that the ability to schedule updates on time in controlled experiments does not guarantee their successful delivery under network disruptions.