An Artificial Circadian System for a Slow and Persistent Robot

As robots become persistent agents in natural, dynamic environments, the ability to understand and predict how that environment changes becomes more valuable. Circadian rhythms inspired this work, demonstrating that many organisms benefit from maintaining simple models of their environments and how they change. In this work, we outline an architecture for an artificial circadian system (ACS) for a robotic agent. This entails two questions: how to model the environment, and how to adapt robot behavior based on those models. Modeling is handled by treating relevant environment states as time series, to build a model and forecast future values of that state. The forecasts are considered special percepts, a prediction of the future state rather than a measurement of the current state. An ethologically-based action-selection model incorporates this knowledge into the agent’s decision making. The approach was tested on a simulated precision agricultural task - pest monitoring with a solar powered robot - where it improved performance and energy management.