Abstract
Dead-reckoning and star-reckoning are two basic self-localization methods for vehicles, but each method has inherent weaknesses. Sensor-fusion via an extended Kalman filter is a suitable method to compensate these weaknesses. However, the extended Kalman filter requires the stochastics of measurement errors for implementation. In this paper, we propose a new sensor fusion method by using an initial state observer. We confirm the effectiveness and usefulness of our method by computer simulation and experiments for self localization of a two-wheeled mobile robot.
