Potential Field Based Swarm Exploration of Processes Governed by Partial Differential Equations

Abstract

In robotic exploration of gas diffusion processes, domain knowledge can improve and speed up estimation of unknown spatial gas concentrations or source localization. Here, a novel robotic swarm exploration strategy for such processes governed by a partial differential equation is presented. A probabilistic process description incorporates physical domain knowledge into the exploration strategy and allows for estimation of concentrations and source distribution, as well as for quantification of the uncertainty of estimated parameters over the spatial exploration domain. Artificial potential fields are used to guide a multi-agent system of non-holonomic robots towards regions of high uncertainty, i.e. informative sampling locations, while avoiding collisions. The developed strategy is evaluated in simulations as well as in Hardware-in-the-Loop experiments. It is shown to outperform a systematic sampling strategy for swarms of the same size.