Balancing Exploration and Exploitation : A Neurally Inspired Mechanism to Learn Sensorimotor Contingencies

Abstract

The learning of sensorimotor contingencies is essential for the development of early cognition. Here, we investigate how such process takes place on a neural level. We propose a theoretical concept for learning sensorimotor contingencies based on motor babbling with a robotic arm and dynamic neural fields. The robot learns to perform sequences of motor commands in order to perceive visual activation from a baby mobile toy. First, the robot explores the different sensorimotor outcomes, then autonomously decides to utilize (or not) the experience already gathered. Moreover, we introduce a neural mechanism inspired by recent neuroscience research that supports the switch between exploration and exploitation. The complete model relies on dynamic field theory, which consists of a set of interconnected dynamical systems. In time, the robot demonstrates a behavior toward the exploitation of previously learned sensorimotor contingencies and thus selecting actions that induce high visual activation.