Improving the control and guidance of unmanned underwater vehicles (UUVs) operating in unkown or dynamic environments is a challenging task, which is faced with lots of uncertainties when it comes to modeling the motion behavior of such robots. Such uncertainties arise from modeling complex physical behaviors due to the interaction between the vehicle’s body and its surrounding environment. The advent of long-term underwater missions in a real world environment imposes complex or unforeseen circumstances that a robot could encounter, and therefore adapting to such situations is an essential aspect of such missions. The goal of this thesis is to develop and test a new intelligent framework that improves the robustness of UUV motion models. By monitoring and extracting features from the onnoard sensory data, the framework is responsible for identifying the mismatches between the model predictions and the outer world, and therefore adapt online the models to the new situations.