The focus of many industrial and research entities on achieving full robotic autonomy increased in the past few years. In order to achieve full robotic autonomy, a fundamental problem is the localization, which is the ability of a mobile platform to determine its position and orientation in the environment. In this thesis, several problems related to the localization of autonomous platforms are addressed, namely, visual odometry accuracy and robustness;​ uncertainty estimation in odometries;​ and accurate multi-sensor fusion-based localization. Beside localization, the control of mobile manipulators is also tackled in this thesis. First, a generic image processing pipeline is proposed which, when integrated with a feature-based Visual Odometry (VO), can enhance robustness, accuracy and reduce the accumulation of errors (drift) in the pose estimation. Afterwards, since odometries (e.g. wheel odometry, LiDAR odometry, or VO) suffer from drift errors due to integration, and because such errors need to be quantified in order to achieve accurate localization through multi-sensor fusion schemes (e.g. extended or unscented kalman filters). A covariance estimation algorithm is proposed, which estimates the uncertainty of odometry measurements using another sensor which does not rely on integration. Furthermore, optimization-based multi-sensor fusion techniques are known to achieve better localization results compared to filtering techniques, but with higher computational cost. Consequently, an efficient and generic multi-sensor fusion scheme, based on Moving Horizon Estimation (MHE), is developed. The proposed multi-sensor fusion scheme:​ is capable of operating with any number of sensors;​ and considers different sensors measurements rates, missing measurements, and outliers. Moreover, the proposed multi-sensor scheme is based on a multi-threading architecture, in order to reduce its computational cost, making it more feasible for practical applications. Finally, the main purpose of achieving accurate localization is navigation. Hence, the last part of this thesis focuses on developing a stabilization controller of a 10-DOF mobile manipulator based on Model Predictive Control (MPC). All of the aforementioned works are validated using numerical simulations;​ real data from:​ EU Long-term Dataset, KITTI Dataset, TUM Dataset;​ and/or experimental sequences using an omni-directional mobile robot. The results show the efficacy and importance of each part of the proposed work.