Anticipating Future Experience using Grounded Sensorimotor Informational Relationships

Abstract

Operational definitions and applications of the sensorimotor experience of an artificial embodied organism are presented along with a mathematical metric for distance between experiences based on Shannon information. We describe a simple robotic experiment that illustrates how an artificial embodied agent can use its own history of experience combined with the experience metric to predict future experience. Present sensorimotor experience is used to find the most similar past experience using the geometry of its growing and changing experience metric space. This is then used to ground the ontogeny of autonomous prospective capability in interacting with the environment, e.g. to anticipate forthcoming changes in environment based on temporally extended past experiences.