Peekaboo: Effect of Experience Length on the Interaction History Driven Ontogeny of a Robot

Abstract

The game peekaboo, ordinarily played between an adult and baby, is used as a situation where a robot may develop social interaction skills such as rhythm, timing and turn taking, using its experience and history of interactions over different temporal horizons. We present experiments using a robot that explore the length of experiences in an architecture that selects action based on a metric space consisting of previous experience and feedback from the environment. Results show that sequences of interactions that allow the robot to play the game successfully emerge from the interplay between environmental or social feedback and experience of various lengths.