Agent Architecture for a Real World Autonomous Virtual Guide: Interaction between the Decision and Perception Processes and Environment Representation

—Museums like marine aquariums are facing a difficult problem when trying to deliver information to their visitors. The exhibits they propose are dynamic by definition. Each may contain multiple autonomous entities that need to be described to the visitor. Classical communication means (panels, audio-guides ...) are static and do not adapt to the constraints of such exhibits. We propose to use an autonomous virtual guide embedded inside the environment in order to describe it to the visitors. To describe this environment and the entities it contains, the agent must be able to perceive it. Doing so is a challenging task due to the dynamic and non-predictable aspects of this environment. In this article, we propose an architecture able to build a partial representation of such an environment that tends to be the " best possible representation " depending on the ongoing task. This is possible by setting up an interaction loop between the perception and decision processes of our intelligent agent. We describe this architecture and provide some results showing how this interaction effectively takes place in an experimental environment and in a real application setting. Index Terms—Vision and scene understanding-architecture and control structure-distributed artificial intelligence-intelligent agents. I INTRODUCTION During the past ten years, museums and related structures (zoo, aquarium...) have evolved from simple artefacts, collections repositories to some sort of cultural attraction and visitor-centered exhibits. This evolution leads them to the installation and setup of new ways to communicate with the visitor [11]. The visitor's behaviour also evolved and nowadays new cultural experiences are expected. The parallel evolution of technologies with computers and multimedia in particular has provided interesting communication means between museums and their visitors. The World Wide Web and multimedia CD-ROMs [4] gave the opportunity to reach more people and to provide interactive multimedia presentations about the museum's exhibits. Portable devices and ubiquitous computing introduced the notion of adaptivity in the information presentation process, allowing that information to be tailored to the visitor's location and/or preferences [24, 22]. Robotics also stepped into the cultural learning world through multiple tour-guide robots [6, 16]. Virtual and augmented reality have also been used as a new way to provide information to the visitor in the real [20] or virtual [28] museum. Despite all these propositions, visitors usually prefer the historical mean of communication in museums: the human guide. This guide accompanies the visitor during their exploration of the exhibit, emphasising …