Social Sensorimotor Contingencies

Abstract: As the field of robotics advances, more robots are employed in our everyday environment. Thus, the implementation of robots that can actively engage in physical collaboration and naturally interact with humans is of high importance. In order to achieve this goal, it is necessary to study human interaction and social cognition and how these aspects can be implemented in robotic agents. The theory of social sensorimotor contingencies hypothesises that many aspects of human-human interaction depend on low-level signalling and mutual prediction. In this thesis, I give an extensive account of these underlying mechanisms and how research in human-robot interaction has incorporated this knowledge. I integrate this work in human-human and human-robot interaction into a coherent framework of social sensorimotor contingencies. Furthermore, I devise a generative model based on low-level latent features that allows inferences about other agent's behaviour. With this simulation experiment I demonstrate that embodied cognition can explain behaviour that is usually interpreted with help of high-level belief and mental state inferences. In conclusion, the implementation of these low-level processes in robots creates a more natural and intuitive interaction without the need of high-level representations.