An interpretable approach to estimate the self-motion in fish-like robots using mode decomposition analysis.

The artificial lateral line system, composed of velocity and pressure sensors, is the sensing system for fish-like robots by mimicking the lateral line system of aquatic organisms. However, accurately estimating the self-motion of the fish-like robot remains challenging due to the complex flow field generated by its movement. In this study, we employ the mode decomposition method to estimate the motion states based on artificial lateral lines for the fish-like robot. We find that primary decomposed modes are strongly correlated with the velocity components and can be interpreted through Lighthill's theoretical pressure model. Moreover, our decomposition analysis indicates the redundancy of the sensor array design, which is verified by further synthetic analysis and explained by flow visualization. Finally, we demonstrate the generalizability of our method by accurately estimating the self-states of the fish-like robot under varying oscillation parameters, analyzing three-dimensional pressure data from the computational fluid dynamics simulations of boxfish (Ostracion cubicus) and eel-like (Anguilla anguilla) models, and robustly estimating the self-velocity in complex flows with vortices caused by a neighboring robot. Our interpretable and generalizable data-driven pipeline could be beneficial in generating hydrodynamic sensing hypotheses in biofluids and enhancing artificial-lateral-line-based perception in autonomous underwater robotics.