Online and robust localization in a large scale outdoor environment is an essential component for self-driving vehicles. This paper addresses the theoretical and experimental development of a 6DOF localization approach with an online Unscented Rauch-Tung-Striebel (RTS) Smoother. This work focuses on the performance evaluation of the Unscented RTS smoother from a low-cost inertial sensor and consumer-grade Differential Global Positioning System (DGPS). The method is evaluated on a publicly available dataset (with centimetre accuracy benchmark) where we compare our results against the stand-alone Unscented Kaiman filter (UKF) and an offline unscented RTS smoother. The extensive evaluation against the conventional approaches demonstrates the effectiveness of our approach, capable of providing accurate/online vehicle's localization information.