Grounded Sensorimotor Interaction Histories in an Information Theoretic Metric Space for Robot Ontogeny

摘要

We motivate and present a definition of an embodied, grounded individual sensorimotor interaction history, which captures the time-extended behavior characteristics of humans and many animals. We present an architecture that connects temporally extended individual experience with capacity for action, whereby a robot can develop over ontogeny through interaction. Central to this is an information theoretic metric space of sensorimotor experience, which is dynamically constructed and reconstructed as the robot acts. We present results of robotic experiments that establish the predictive efficacy of the space and we show the robot developing the capacity to play the simple interaction game “peekaboo.” A quantitative investigation of the appropriate horizon length of experience for the game reveals the relationship between the length of experience and the cycle time of interaction, and suggests the importance of multiple, and possibly self-adaptive, horizon lengths.