Sonar interpretation using regions of constant depth (RCDs)

Describes an approach to the automatic interpretation of sonar data for mobile robot navigation. Based on previous work and experimental evidence, the author believes that the most natural and reliable features which are contained in sonar scans are not straight line segments but circular arcs; he refers to these features as regions of constant depth. The interpretation of sonar data for navigation involves two tasks: the prediction of expected RCDs for known geometric features, and the explanation of unexpected RCDs corresponding to unknown features of the environment. Algorithms are presented for predicting RCDs for plane, cylinder, corner and edge features, and for extracting RCDs from observed sonar data. Experimental results in classifying unexpected RCDs for map building using the standard Polaroid ultrasonic ranging system are presented