Autonomous monitoring and surveillance systems are increasingly critical for indoor security and facility management. Historically these applications have relied on static, rigid and deterministic algorithms. Traditional robotic perception utilizes standard object detection models that, while capable of identifying bounding boxes, lack true semantic understanding and contextual reasoning. Furthermore, classic systems struggle to correlate high-level visual identity with precise low-level spatial data, limiting their ability to react intelligently to dynamic, unstructured environments. The integration of Large Language Models (LLMs) and Vision-Language Models (VLMs) into robotic frameworks represents a significant milestone in overcoming these limitations, shifting the paradigm from programmatic execution to semantic situational awareness.