Robot Jacobian control: a new approach via artificial neural networks

A new approach in applying the theory of cognition is presented, where the concepts of artificial neural networks are combined with conventional robot control theory to produce a massively-parallel adaptive controller. The contribution given herein is two folds. First, a parallel structure of a semi-symbolic representation of the equations is presented, where the computational burden is cut down. Second, certain concepts of the theory of cognition are employed in the design of a multi-layered neural network, in which adaptation for any changes in the robot model or the environment can be accommodated for via the back-propagation of errors throughout the network. To illustrate the validity of the presented algorithm, simulation results are reported for the Unimation PUMA 560 manipulator with 6 degrees-of-freedom