Bayesian gas source localization and exploration with a multi-robot system using partial differential equation based modeling

Model based approaches, such as those that use partial differential equations (PDE), lend themselves to gas distribution mapping and gas source localization. Moreover, they also permit constructing intelligent sampling strategies. However, a realistic mathematical model of gas dispersion is complex and computationally expensive to solve. This is especially the case for inverse problems, where sources are estimated based on concentration measurements. In this paper, we propose a probabilistic model based on a diffusion PDE to approximate the complex behavior of gas dispersion. This model is used (i) to identify the sources, using ideas from Sparse Bayesian Learning, and (ii) to guide a multi-agent robotic system to measurement locations, which assists the source localization. The potential of the approach is shown in experiments, where laminar gas plumes are simulated using an open-source CFD-based filament gas dispersion simulator. The exploration is carried out using multiple real robots implementing the proposed algorithm