The research embodies a novel approach to visually guided control of multijoint robot movement under conditions of unforeseen errors. It is based on a new theory of mass-acting neuron models integrating visual sensing and motor manipulation. The theory uses principles of neural network organization to derive parallel architectures and algorithms for hierarchies of computations, which will implement libraries of basic motions. The theory has been well-developed in a single-joint case, that of ballistic eye movements. The computer simulations will be used to explore extension of the theory to the case of a multi-joint robot arm.