Model based object localization and shape estimation using electric sense on underwater robots

Abstract

Recently, biologists have shown that the weakly electric fish are able to estimate the electric nature, the localization and the 3D geometric properties of an object using active electric sense. Incredibly, the Gnathonemus petersii performs this task in the dark only by moving towards and around the object, its vision is not required. In this paper, we proposed to address the challenging issue of object localization and shape estimation using a real underwater robot equipped with artificial electric sense. To that end, we used a corrected version of the dipolar tensor dedicated to small objects [Ammari et al., 2014] able to capture the electric response of big objects (typically objects whose size is about the one half of the robot length < 10cm). The principal contribution consists in the development of a multi-scale exhaustive search algorithm based on this tensor that allows to estimate in a same step the localization, orientation and shape of an object from electric currents measured along a given trajectory close to the object. Over 108 experiments, our method shows good results as on average we obtained 18% of shape error, 25° of orientation error and 1cm of localization error within a range of [5, 11]cm distance with the robot. These results are promising since the problem solved is known to be complex localization and shape being intricately linked in the electrical measurements [Rasnow, 1996].