A neural network for object recognition through sonar on a mobile robot

Abstract

A novel approach to sonar-based object recognition for use on an autonomous robot is described. An inexpensive Polaroid sonar, coupled with an A/D data acquisition board installed in a LINUX-based PC, provides full-echo sonar information to a fuzzy ARTMAP neural network classification system. Initial experimental results are described which illustrate recognition of five objects and generalization across distance. Findings suggest that typical indoor objects can be distinguished over a range of distances with high accuracy solely on the basis of information contained in the ultrasonic sonar echoes.