Sampling and Exploration Strategies for Multi-Agent Seismic Imaging Using 3D Full Waveform Inversion

For future planetary missions, we envision the use of multi-agent networks that autonomously and cooperatively explore a planet's subsurface. To this end, we consider seismic imaging techniques such as full waveform inversion to obtain tomographic images of the subsurface. However, seismic imaging techniques have not been largely investigated yet within the context of robotic exploration. Therefore, in this work we investigate full waveform inversion for subsurface exploration by a multi-agent network. To this end, we assume a small robotic network where each agent is equipped with a geophone. The network shall perform 3D imaging based on measurements recorded over the subsurface of interest. We investigate different sampling and exploration strategies to explore a synthetic subsurface model and evaluate the strategies with respect to their resulting imaging performance. Based on 3D numerical simulations we show that a topology with random agent positions obtains better imaging performance compared to a uniform grid formation. Furthermore, separating the exploration area into smaller areas that overlap with each other gives further performance gains. Our work provides insights in how full waveform inversion can be employed for subsurface exploration by a multi-agent network.