Spoton: Just-in-time active event detection on energy autonomous sensing systems

Abstract

We propose SpotON, which is an active event detection system that runs on harvested energy and adapts its sleeping cycle to match the distribution of the arrival of the events of interest. Existing energy harvesting systems wake up periodically at a fixed rate to sense and process the data to determine if the event of interest is happening. In contrast, SpotON employs reinforcement learning to learn the pattern of events at run-time and uses that knowledge to wake itself up when events are most likely to happen. Being able to remain asleep more often than a fixed wake-up system, SpotON is able to reduce energy waste, increase the amount of harvested energy, and be able to remain active for longer period in time when the events of interest are more likely to occur. We conduct a simulation-driven experiment to compare our proposed solution with a fixed-schedule system and results show that SpotON is able to capture 2–5X times more events and is 3–12X more energy-efficient than the baseline.