Robot Navigation Using Biosonar for Natural Landmark Tracking

Abstract

A biosonar based mobile robot navigation system is presented for the natural landmark classification using acoustic image matching. The aim of this approach is to take advantage of the perceived properties of bats' prey and landmark identification mechanisms for mobile robots' tracking of natural landmarks. Recognizing natural landmarks like trees through sequential echolocation and acoustic image analyzing allows mobile robot to update its location in the natural environment. In this work, a working implementation of the biosonar system on a mobile robot is shown. It collects sequential echoes to produce acoustic images through digital signal processing (DSP), then compresses images with discrete cosine transform or pyramid algorithm. Fast normalized cross correlation (FNCC) and kernel principal component analysis (KPCA) are respectively used to make the final classification. Experimental result indicates that a mobile robot can achieve the ability of natural landmark classification only based on biomemetic sonar, the topological congruency of the relational structure with cross correlation in acoustic images is reliable in time domain, while the kernel principle component analysis based classification is robust in frequency domain and demands fewer echolocation for landmark classification.