Learning Decision Trees for Mapping the Local Environment in Mobile Robot Navigation

摘要

This paper describes the use of the C4.5 decision tree learning algorithm in the design of a classifier for a new approach to the mapping of a mobile robot's local environment. The decision tree uses the features from the echoes of an ultrasonic array mounted on the robot to classify the contours of its local environment. The contours are classified into a finite number of two dimensional shapes to form a primitive map which is to be used for navigation. The nature of the problem, noise and the practical timing constraints, distinguishes it from those typically used in machine learning applications and highlights some of the advantages of decision tree learning in robotic applications. 1 INTRODUCTION Ultrasonic sensors are often used to measure distances and in the majority of cases the time of flight of an ultrasonic pulse of energy is utilized. That is, the total time taken for the pulse to leave the transmitter, reach the object in question, and then on the final leg of its journey...