The recent advancements in Vision-Language Models (VLMs) and their applications across various domains, particularly in navigation and autonomous systems, are significantly reshaping the landscape of research. A notable trend is the shift towards end-to-end navigation frameworks that leverage VLMs for direct action selection, bypassing traditional perception-planning-control pipelines. This approach not only simplifies the navigation process but also enhances its generalizability and adaptability to diverse tasks. Additionally, there is a growing emphasis on integrating multimodal data, such as object detection and urban space representations, to improve the robustness and interpretability of autonomous systems. The use of foundation models for semantic enhancement in SLAM systems is also gaining traction, offering more precise object-level mapping and dynamic scene updates. Furthermore, the development of probabilistic frameworks for instance-aware semantic mapping introduces a new dimension of uncertainty quantification, which is crucial for enhancing the reliability of robotic systems in complex environments. These innovations collectively push the boundaries of what is possible in embodied intelligence and autonomous navigation, paving the way for more sophisticated and adaptable robotic applications.