Abstract: In recent years, deep learning techniques have made significant progress in the field of Simultaneous Localization and Mapping (SLAM) for mobile robots, providing new ideas to address the challenges faced by traditional visual SLAM in dynamic environments. Firstly, this paper summarises the limitations of traditional visual SLAM in terms of pre-processing, visual odometry and loop-closure detection modules. Subsequently, this paper focuses on the application of deep learning in visual SLAM, highlighting the deep learning-based preprocessing, visual odometry, and loop-closure detection modules and how they can improve the robustness and accuracy of visual SLAM. Finally, this paper also discusses the challenges faced by deep learning-based SLAM and looks forward to future research directions, including lightweight network design, long-term modelling of scenes, and self-supervised learning, in order to promote deep learning SLAM in practical applications.