Route learning in mobile robots through self-organisation

Abstract

This paper describes route-learning experiments with an autonomous mobile robot in which map building is achieved through a process of unsupervised clustering of sensory data. The resulting topological mapping of the robot's perceptual space is used for subsequent navigation tasks such as route following. After the autonomous mapbuilding process is completed, the acquired generalised perceptions are associated with motor actions, enabling the robot to follow routes autonomously. The navigation system has been tested extensively on a Nomad 200 mobile robot, it is reliable and copes with noise and variation inherent in the environment. One important aspect of the map building and route following system described here is that relevance or irrelevance of perceptual features is determined autonomously by the robot, not through predefinition by the designer. Secondly, the presented route learning system enables the robot to use the map for association of perception with action, rather than localisation alone.