Recurrent neural network for ballistic actions: A study with the iCub

This work aims to design an artificial neural network architecture to efficiently learn a model for the offline control of robot movements, with the final goal is the derivation of engineering principles for the development of autonomous systems that are capable to refine their motor skills in an open-ended process. In particular in this abstract we will show results of experiments with the humanoid robotic platform iCub that execute a task with ballistic movements. Ballistic movements are made without the online use of sensory feedback evolving during the action. All movements in a ballistic action are ballistic. The neural system that controls the robot is a three layer recurrent neural network.