Interactive Learning of Inverse Kinematics with Null-space Constraints using Recurrent Neural Networks

Abstract

Industrial co-worker scenarios require a save, flexible, and efficient control of robots. Our cognitive system FlexIRob as a prototype for human robot interaction in industry allows flexible handling and fast reconfiguration of a compliant redundant robot system by use of a machine learning approach. Problem Statement: Learning inverse kinematics with redundancy resolution in physical human robot interaction.