Underwater SLAM with ICP Localization and Neural Network Objects Classification

The aim of this paper is to propose a technique for Simultaneous Localization and Mapping in underwater environments by means of acoustic sensors. The proposed procedure consists in the application of suitable Neural Network and Iterative Closest Point algorithms for objects detection, agent localization and map construction. General Regression Neural Network and improved ICP algorithms are implemented in order to process sonar data, to minimize the computational time and to maximize efficiency in localization tasks without using dynamical models of the agent. Experimental tests have been performed in a simple, structured static environment collecting data by means of a single-beam, mechanically scanning sonar. Results show good performances of the procedures in simple but meaningful situations.