Motion control of two-link flexible-joint robot with actuator nonlinearities, using neural networks and direct method

We present a state-feedback control of a two-link flexible-joint robot. We use three-layer neural networks to learn the unknown parts of the desired control laws. Therefore, the control algorithm does not require the mathematical model representing the robot. The neural networks´ weights are adapted on-line. Therefore, the control system can adapt to changes during operation such as payload changes. We use a smooth variable structure controller to handle uncertainties from the neural network approximation, external disturbances, deadzone, and backlash. In our simulation, we obtain the actual plant model from experiments