AI Agents for Evaluating Universal Design in XR Architectural and Heritage Environments

Abstract. Evaluating spatial accessibility and universal design in architectural environments typically requires extensive user
testing with diverse participant groups, which can be dicult to conduct during early design stages. This challenge is
particularly relevant in museum and cultural heritage environments, where historical buildings must often be adapted to
contemporary accessibility standards. This paper explores how Extended Reality (XR) environments combined with AI-driven
agents may support exploratory evaluation of accessibility conditions in architectural models. Building on previous research with the Virtual Scenario Responder (VSR), the proposed framework integrates behavioral data capture, agent-based exploration, and multi-modal computer vision analysis within a spatially anchored digital twin. Human behavioral data collected in real environments are used to seed agent simulations that navigate architectural models and identify potential Areas of Concern (AOCs) related to accessibility. Preliminary experiments suggest that combining human behavioral traces with
automated agent exploration may provide a promising method for identifying spatial accessibility challenges during early design stages. Such approaches may be particularly relevant for museum and heritage environments, where XR technologies are increasingly used to explore visitor experience, accessibility, and inclusive spatial interpretation.