Multiple Representations for Mobile Robot Vision

Abstract

The Autonomous Land Vehicle (ALV) must be able to navigate over a wide variety of terrain, using its visual sensors to recognize objects and correct its path. It can rely on three important resources for information about its environment: a time-of-flight range sensor, an inertial guidance system that provides accurate dead-reckoning information, and a knowledge base that specifies the locations, shapes, and appearances of objects in the world. This paper addresses the problems in integrating these resources into a closed-loop control system that can adaptively correct the ALV's route through the environment.