Intelligent control with the growing competitive linear local mapping neural network for robotic hand-eye coordination

Several intelligent features are embedded in the growing competitive linear local mapping neural network. They result in an adaptive, fast-learning, very efficient control scheme, suited for closed-loop systems. The use of the Moore-Penrose pseudoinverse allows us to generalize our results to nonredundant manipulators and to redundant visual features. Information acquired during the current movement can update the control matrix. Instead of learning the entire workspace, only one or a few trajectories can be learnt, in one pass per trajectory. The feedback loop and the fast convergence of RLS allow an active learning of the local Jacobians