Belief formation by constructing models

Abstract

The notion of uncertainty in robotics has to date largely involved the uncertainty, or variability, in the information the robot was receiving from its sensors, information about an outside world which itself is known, certain, unambiguous. For example, the uncertainty the robot may experience in answering the question “Is that object a cube?”. With the emergence of autonomous, ‘really useful robots’, however, robots also need to address the problems of uncertainty which will require the robot to reason about the real world, “It is a cube, but is there something hiding behind it?”. In this paper we reconsider a mode of ‘intelligent’ uncertain reasoning, based on forming beliefs by constructing models, which would seem to be a natural, and simple, step in the evolution of certain of today's robots and may provide the robot with a means of reasoning about the uncertainties of the world it encounters.

Supported by a Science & Engineering Research Council Research Assistantship.