ABSTRACT
The success of robotic assembly for odd form electronic components depends heavily on the ability to monitor and control the insertion force. For this purpose, a joint disturbance observer can be applied. However the observed disturbance signal includes the effects of model uncertainties. An approach of using a neural network to learn the parametric and unstructured uncertainties in robot manipulators is proposed. Furthermore a true teaching signal for learning the uncertainties is obtained. After learning, the neural network is embedded in the structure of the joint torque disturbance observer to compensate for the uncertainties in the robot dynamic model. As the result, accurate estimate of the external disturbance force can be deduced.
