Robot Identification using Dynamical Neural Networks

The authors solve the identification problem of a robotic manipulator using dynamical neural networks. They propose a dynamical backpropagation scheme that can learn and identify nonlinear systems without needing any prior knowledge about the system to be identified. Simulations show that the proposed algorithm can handle abrupt changes in input data, that the error converges quickly to zero, and that the network can effectively perform after the training stops, even when the input waveforms have not been previously presented. >