Neural-based underwater spherical object localization through electrolocation

Abstract

Navigation of cluttered underwater environments remains to this day a challenging task in mobile robotics. Applying an electric field to a mobile robot's direct environment and measuring perturbations of this field, one is able to detect the presence of foreign objects in close proximity of the system. In addition, one is also able to infer a range of information relative to the detected objects, such as their position or electrical characteristics. Extracting such information from available measures typically requires a model (analytical, numerical or heuristic) descriptive of the relationship from geometry of the scene to measures performed (typically referred to as forward model), or of the inverse relationship (inverse model). In the following, we directly extract one such model from experimental data, and capture a forward model using a neural formalism. Then, using an iterative procedure, we are able to estimate the position of a detected object and assess the degree of confidence one can place on this estimate. Merit of the approach is illustrated using experimental data for a spherical object.