Lazy acquisition of place knowledge

Abstract

In this paper we define the task of place learning and describe one approach to this problem. Our framework represents distinct places as evidence grids, a probabilistic description of occupancy. Place recognition relies on nearest neighbor classification, augmented by a registration process to correct for translational differences between the two grids. The learning mechanism is lazy in that it involves the simple storage of inferred evidence grids. Experimental studies with physical and simulated robots suggest that this approach improves place recognition with experience, that it can handle significant sensor noise, that it benefits from improved quality in stored cases, and that it scales well to environments with many distinct places. Additional studies suggest that using historical information about the robot‘s path through the environment can actually reduce recognition accuracy. Previous researchers have studied evidence grids and place learning, but they have not combined these two powerful concepts, nor have they used systematic experimentation to evaluate their methods‘ abilities.