UC Riverside

The emergence of intermittently-powered batteryless devices (IPDs) for energy harvesting in harsh environments has spurred research into addressing their challenges. This dissertation integrates and extends three key contributions in the realm of real-time sensing and task scheduling for such devices. First, a scheduling framework prioritizing atomic sensing operations for periodic execution on IPDs is proposed, ensuring efficient utilization of intermittent power sources and verifying task schedulability under varied charging conditions. Secondly, a novel energy-adaptive real-time sensing framework is introduced, leveraging lightweight machine learning to predict energy availability and adjust task schedules to ensure continuous sensing operations while meeting data freshness requirements. Thirdly, CARTOS, a charging-aware real-time operating system for IPDs, is presented, employing mixed-preemption scheduling and just-in-time checkpointing to ensure both computational and peripheral tasks execute efficiently amidst energy variability while supporting processing chains with adaptable scheduling.These contributions collectively address the challenges of ensuring continuous operation, data freshness, and reliable execution on IPDs in IoT applications. Evaluations across real hardware experiments and simulations demonstrate the superiority of the proposed frameworks over existing methods. Overall, these advancements offer practical solutions for developing resilient real-time sensing applications on batteryless devices, paving the way for their widespread deployment in diverse scenarios.