A multilayer neural net (NN) controller for a general serial-link robo
t arm is developed. The structure of the NN controller is derived usin
g a filtered error approach. It is argued that standard backpropagatio
n tuning, when used for real-time closed-loop control, can yield unbou
nded NN weights if: (1) the net can not exactly reconstruct a certain
required control function, (2) there are bounded unknown disturbances
in the robot dynamics, or (3) the robot arm has more than one link (i.
e. nonlinear case). On-line weight tuning algorithms including correct
ion terms to backpropagation, plus an added robustifying signal, guara
ntee tracking as well as bounded weights. The correction terms involve
a second-order forward-propagated wave in the backprop network.