Persistent Preservation of a Spatio-temporal Environment Under Uncertainty

Abstract

This paper tackles the spatio-temporal areas restoration problem for a single robot when faced with state uncertainty: a robot, with limited battery life, deployed in a known environment, persistently plans a schedule to visit areas of interest and charge its battery as needed. The temporal properties of areas decay over time, wherein the decay is only partially observable and evolves over time, potentially with correlation among areas. The goal is to restore the temporal properties so that the time the measured property values are below a certain threshold is minimized.Our previous work formulated the spatio-temporal areas restoration problem assuming that the decays are known. Instead, in this paper, we relax that assumption and account for the uncertainty, proposing a heuristic to measure the discounted opportunity cost of a visit, which induces risk-aversion to revisit overlooked areas, and adding a component that learns the decay parameters in each area as well as potential correlation among areas. The learning component can then be used to predict future trends and be incorporated in the heuristic forecast. Moreover, the algorithm learns and constantly adjusts for noise that can happen during mission. We show in experiments using a robotics simulator that our devised approach is able to maintain areas above the critical threshold better than existing state-of-the-art methods from related problems. This contribution enables a robot to come up with an effective schedule efficiently for preserving spatio-temporal properties of an environment considering realistic scenarios–which has markedly impact in important environmental applications.