Mobile robot self-localisation and measurement of performance in middle-scale environments

Abstract This paper addresses the question of self-localisation in autonomous mobile robot navigation, i.e., the task of indentifying places after previous exploration and map building by the robot. We present a novel localisation system which accumulates both exteroceptive and proprioceptive sensory evidence over time to localise, without requiring prior knowledge of the robot's position. We show that the system relocalises successfully on a real robot in middle-scale environments containing transient changes such as moving people. In addition, a general performance metric and a standard experimental procedure are introduced, allowing disparate localisation systems to be compared on the same robot in the same environment. To demonstrate the utility of the approach taken, we test the evidence-based localisation system in six different environments, comparing its performance to that of localisation using dead reckoning or currently observable landmarks alone. In addition, the results provide us with some useful quantitative measures for characterising different environments.