Abstract

One of the key challenges of current state-of-the-art robotic deployments in public spaces, where the robot is supposed to interact with humans, is the generation of behaviors that are engaging for the users. Eliciting engagement during an interaction, and maintaining it after the initial phase of the interaction, is still an issue to be overcome. There is evidence that engagement in learning activities is higher in the presence of a robot, particularly if novel [1], but after the initial engagement state, long and non-interactive behaviors are detrimental to the continued engagement of the users [5, 16]. Overcoming this limitation requires to design robots with enhanced social abilities that go past monolithic behaviours and introduces in-situ learning and adaptation to the specific users and situations. To do so, the robot must have the ability to perceive the state of the humans participating in the interaction and use this feedback for the selection of its own actions over time [27].