An Empirical Evaluation of Context-Sensitive Pose Estimators in an Urban Outdoor Environment

When a mobile robot is executing a navigational task in an urban outdoor environment, accurate localization information is often essential. The difficulty of this task is compounded by sensor drop-out and the presence of non-linear error sources over the span of the mission. We have observed that certain motions of the robot and environmental conditions affect pose sensors in different ways. In this paper, we propose a computational method for localization that systematically integrates and evaluates contextual information that affects the quality of sensors, and utilize the information in order to improve the output of sensor fusion. Our method was evaluated in comparison with conventional probabilistic localization methods (namely, the extended Kalman filter and Monte Carlo localization) in a set of outdoor experiments. The results of the experiment are also reported in this paper