Reach and Grasp for an Anthropomorphic Robotic System based on Sensorimotor Learning

In this article, we present a neurobiologically inspired multinetwork architecture based on knowledge of cortico-cortical connectivity and its application on an anthropomorphic head-arm-hand robotic system to provide reach-and-grasp kinematics based on multimodal sensorimotor learning. The system incorporates artificial neural network modules (matching units) trained by the locally weighted projection regression (LWPR) algorithm that enables progressive learning from simple to more complex sensorimotor tasks. We report the actual performance of the system by comparing the simulation with the experimental results obtained by the implementation on the real world artefact