Perceptual-motor sequence learning via human-robot interaction

Abstract

The current research provides results from three experiments on the ability of a mobile robot to acquire new behaviors based on the integration of guidance from a human user and its own internal representation of the resulting perceptual and motor events. The robot learns to associate perceptual state changes with the conditional initiation and cessation of primitive motor behaviors. After several training trials, the system learns to ignore irrelevant perceptual factors, resulting in a robust representation of complex behaviors that require conditional execution based on dynamically changing perceptual states. Three experiments demonstrate the robustness of this approach in learning composite perceptual-motor behavioral sequences of varying complexity. © Springer-Verlag Berlin Heidelberg 2006.