Dynamic recurrent neural networks for modeling flexible robot dynamics

Abstract

The identification of a general class of multi-input and multi-output (MIMO) discrete-time nonlinear systems expressed in the state space form is studied using dynamic recurrent neural network (DRNN) approach. A novel discrete-time DRNN, which is represented by a set of parameterized nonlinear difference equations and has the universal approximation capability, is proposed for modeling unknown discrete-time nonlinear systems. Dynamic backpropagation learning algorithm is discussed extensively in order to carry out the modeling task using the input-output data. A simulation example of modeling flexible robot dynamics is provided to demonstrate the usefulness of the proposed technique.