Active sensing for mobile robots

Abstract

Whilst research work on single sensor modalities has tended to concentrate on What is it? perception, the need for active, mobile robots is more typically Where is it? perception. For obstacle avoidance, local planning, hand-eye coordination or simple sensor skills, the need is usually that of knowing what parts of the world are occupied rather than the later, more sophisticated task of knowing what is doing the occupying. This paper outlines a series of experiments showing how simple sensor reflexes can be developed, integrated and used as the basis for where-is-it world modelling. Examples are given for optical vergence and shape-from-stereo as well as showing how such information can be used directly for clear-path local navigation. >