This thesis explores advancements in Simultaneous Localization and Mapping (SLAM) for autonomous ground vehicles, focusing on the integration of neural networks to enhance localization accuracy and generalizability. The research addresses key challenges in SLAM, including scale drift, environmental adaptability, and computational efficiency. Through a systematic literature review and empirical studies, the thesis evaluates the performance of neural network-based SLAM techniques, particularly in diverse and dynamic environments. The findings highlight the potential of neural networks to improve SLAM by leveraging large, diverse datasets and advanced image enhancement methods. Additionally, the research investigates sensor fusion techniques, combining visual and inertial data to enhance localization performance. The contributions of this thesis provide a comprehensive framework for future research in SLAM-based localization, aimed at improving the generalizability and computational efficiency of autonomous navigation systems.