Learning the expected utility of sensors and algorithms

Abstract

A method is proposed which estimates the expected utility of a sensor being used in a sensor fusion framework. The resulting values are used to predict the subset of sensors which should be read to minimize the total cost of an observation cycle. Preliminary results from experiments taken with three sensors mounted on a mobile robot indicate that the method is indeed capable of reducing the average cost of an observation cycle, and that it is also capable of dynamically tracking conditions which change the expected utility values.<>