It is difficult to represent the nonlinear characteristics in the dyna
mics of robot manipulators by means of a mathematical model. An altern
ative approach of using a neural network to learn the parametric and u
nstructured uncertainties in robot manipulators is proposed. It is the
n embedded in the structure of a joint torque perturbation observer to
compensate for uncertainties in the robot dynamic model. As the resul
t, an accurate estimate of the joint reaction torque against the envir
onment can be deduced. The approach is applied to monitor the insertio
n force during electronic components assembly using a SCARA robot. A t
rue teaching signal of neural network for learning the model uncertain
ties is obtained. Furthermore, a special motion test is conducted to g
enerate the required training data set. After learning, the neural net
work is capable of reproducing the training data. The generalizing abi
lity of the network enables it to output the correct compensation sign
al for a trajectory which it has not been trained. With the proposed t
echnique, it is possible to verify the success of component insertion
in real time and avoid causing damages to the electronic components.