Incremental pose estimation for mobile robots within curvilinear environments

Abstract

This paper addresses the self-localization of a mobile robot, equipped with a laser range scanner, navigating within a curvilinear environment. Specifically, we present a technique for updating a current pose estimate (e.g., deriving from dead-reckoning) by means of a single range measure. Current approaches to this problem, based on the Extended Kalman (EK) estimator, often yield unreliable results due to linearization errors, inherent in curvilinear environments. Therefore, a second-order method is proposed, in which the measurement result is approximated in terms of a two-degree function of the pose estimate error (instead of a linear one). Experiments conducted on a real robot navigating within a curvilinear environment are reported, showing that the errors in the updated estimate, provided by the proposed method, are significantly lower than the errors in the EK estimate.