In this article, an approach to improving the: performance of multi-manipul
ator systems using neural network is presented. This approach is formulated
in the constrained motion framework, within which a nominal feedback contr
ol augmented by a neural network is derived It is shown that the closed-loo
p system with the neural network learning on-line is stable in the sense th
at all signals in the systems are bounded. It is further proved that the pe
rformance of the multi-manipulator system is improved in the sense that the
"size" of a certain error measure decreases as the learning process of the
neural network is iterated. Results of computer simulations conducted to v
erify the analytical conclusions are presented. The results of this work su
ggest that neural networks could be used as "add-on" control modules to imp
rove the performance of industrial robots in execution of tasks involving t
wo or more cooperative manipulators.