Multiple Embodiments For Robots In Heritage Applications Amol Deshmukh, Ruth Aylett, Michael Kriegel and Patricia A. Vargas School Of Mathematical and Computer Science, Heriot-Watt University, Edinburgh EH14 4AS, UK Abstract. In this paper we discuss how one might use multiple em- bodiments to overcome the limitations of specific robot embodiments in heritage applications. We outline our work on migration of an intelligent agent between different embodiments, graphical, mobile and robotic in the context of long-lived robot companions. We summarise CMION, an open source architecture for coordinating the sensors and effectors of such an agent with its mind (its high level decision making processes). Both social robots and embodied conversational characters (ECAs) are em- bodied, the former physically and the later graphically. Physical embodiment has many advantages in user-interaction arising from a shared physical space and a higher degree of presence and engagement. However, it also raises still unsolved engineering problems of power sources, mobility and localization that typically limit the ability of robots to accompany humans, for example across rough ground in an outdoor heritage site. Virtual embodiments are much more transportable but by their nature do not support physical task execution or the degree of social presence needed, for instance, to escort a group of visitors and point at specific aspects of a heritage site. Different physical embodiments may also have different capabilities: one may be small enough to run underneath an arch, and another may not; one may have a manipulator suitable for fine manipulation and pointing; another, one that can lift heavy artefacts. For these reasons, migration, or the ability of a synthetic guide to move from one embodiment to another [1], seems a valuable capacity for robots in heritage applications. We define the migrating element as the robot’s identity, by which we mean those features that persist and make it unique and recognizable from the user’s perspective. These features are themselves a research topic [2] they may include not only common attributes of the different embodiments, for example a similar facial appearance, but also common aspects of interactional behaviour, such as emotional expressiveness, as well as a memory of events and interactions that have taken place in multiple embodiments. We have studied how far users accept the persistence of identity across different embodiments and in what way this changes their interactional behaviour [3] as well as how this impacts the design of migration.