A neurorobotic model of learning to shake a rattle

Abstract

Reward-modulated Hebbian learning is a biologically plausible neural learning mechanism that has been previously applied to a variety of learning tasks. For example, recent work used reward-modulated spike timing dependent plasticity (STDP) to help explain how infants learn to produce syllabic babbling [1]. This project attempts to extend this learning mechanism to a new domain of infant motor development, shaking a rattle. The experiment transduces neural spike trains to adjust frequency of sinusoidal movement around a robotic arm's articulation point. Reinforcement given when the volume, defined as the root mean square (RMS) amplitude, of sound made by a rattle attached to the robot arm exceeded the mean RMS of recent trials.