A neural network approach for robust dynamical control of robot manipulators

A new algorithm for adaptive dynamical control of robot manipulators is proposed. It utilizes a reduced order parameterization together with a computed torque control policy. The teaching signal correlates the total torque with the feedback error. The algorithm was simulated for a three revolute joint robot and various factors such as noisy joint velocity measurements, weaker control, persistency of excitation and choice of learning constants are examined. Implementation issues for parallel processing are discussed.<>