A Computed Torque Controller for Robotic Manipulators Using Nonlinear Neural Network

In robotic control engineering, the conventional computed-torque control algorithm is a simple method to control robots achieving the desired quality by using model parameters including internal dynamics and external disturbances to establish the control law. Practical applicability of this method is normally low since it is difficult to accurately determine such the parameters. In this paper, we propose an intelligent computed-torque control approach for tracking control problems of robotic systems. A neural network structure is first employed for online estimation of the system dynamics in which the learning process is stimulated by a nonlinear mapping function of the control error. From there, the computed-torque control signal is then synthesized using the estimation result and a proportional-derivative control term to result in expected control performance. Stability of the closed-loop system is maintained by Lyapunov analyses. Effectiveness of the proposed control method is extensively verified through intensive simulation results.