A Unified Framework for AR Indoor Navigation and Conversational Information Retrieval in Smart Campus Environments

Abstract

Navigating complex indoor environments like university campuses poses a significant challenge, compounded by the difficulty of accessing specific institutional information. While vision-based Simultaneous Localization and Mapping (SLAM) offers continuous tracking, it suffers from cumulative drift. Sim- ilarly, standard Retrieval-Augmented Generation (RAG) systems for question-answering are prone to context collision. This paper presents a unified mobile application that integrates a robust AR indoor navigation module with an intelligent conversational assistant to address both challenges simultaneously. The pro- posed navigation system fuses the continuous motion tracking of SLAM with high-precision QR code markers for drift correction, providing accurate, real-time visual guidance. For information retrieval, a novel RAG architecture is proposed featuring an LLM-powered router that intelligently directs user queries to one of several specialized vector stores, effectively eliminating con- text collision and improving contextual accuracy. Experimental results confirm that the navigation module provides robust and accurate visual guidance in real-world scenarios. Furthermore, the chatbot's router-based architecture demonstrates a marked improvement in retrieval accuracy, consistently identifying more relevant information than a standard RAG system. The proposed system demonstrates the feasibility of synergistically combining spatial computing with advanced conversational AI to create a comprehensive smart campus assistant, enhancing the student experience by providing seamless access to both physical and digital resources.