Self-organizing neurocontrol

Self-organizing neural network solutions to control problems are described. Competitive networks create spatial filters and geometry connections in a self-organizing fashion. The goal position, the obstacles and the object under control all create neural activities through the filters. Spreading activation that discriminates between the controlled object, the goal position and the obstacles is utilized on the internal representation. A local self-training method and Hebbian learning develop the self-organizing control connections. The algorithm provides manoeuvring capability in unseen scenes.<
>